<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>AI Stack Crunch: New this week</title><link>https://aistackcrunch.com/new/</link><atom:link href="https://aistackcrunch.com/new/feed.xml" rel="self" type="application/rss+xml"/><description>Verified new AI tool and model releases, each with a short honest summary. Becomes The Crunch newsletter every Tuesday.</description><language>en</language><lastBuildDate>Mon, 16 Nov 2026 12:00:00 +0000</lastBuildDate><item><title>Qwen4-Max fully launches after the Nov 11 preview</title><link>https://x.com/Alibaba_Qwen/status/2122148127520289151</link><guid isPermaLink="false">2026-11-16|https://x.com/Alibaba_Qwen/status/2122148127520289151</guid><pubDate>Mon, 16 Nov 2026 12:00:00 +0000</pubDate><source url="https://aistackcrunch.com/new/feed.xml">AI Stack Crunch</source><description>On Nov 16, Alibaba fully launched Qwen4-Max, following the Nov 11 preview of Qwen4-Max-Preview on Alibaba&#x27;s Token Plan. The preview docs listed a 1M-token context window and up to 393,216 output tokens (shared by thinking and the final answer), with thinking on by default, plus image and video understanding, tool calling and structured outputs. Official posts also point to access on QwenWork, Qoder, the Qwen app and QwenStudio. Reports say the Max model&#x27;s weights are closed; smaller Qwen4 models are reported to be promised as open weights later. (Source: Alibaba Cloud)</description></item><item><title>ChatGPT Wallet lets Dots agents make purchases you approve with biometrics</title><link>https://openai.com/index/introducing-chatgpt-wallet/</link><guid isPermaLink="false">2026-11-13|https://openai.com/index/introducing-chatgpt-wallet/</guid><pubDate>Fri, 13 Nov 2026 12:00:00 +0000</pubDate><source url="https://aistackcrunch.com/new/feed.xml">AI Stack Crunch</source><description>OpenAI launched ChatGPT Wallet on Nov 13. ChatGPT Dots agents can make purchases using a card token, and you approve each one with biometrics. It is available to US ChatGPT Pro users only for now, and is built with American Express, Mastercard and Visa. (Source: OpenAI)</description></item><item><title>Google releases Gemini 4 Hydrogen, a faster, lower-cost model for coding and agents</title><link>https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-4-hydrogen/</link><guid isPermaLink="false">2026-11-12|https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-4-hydrogen/</guid><pubDate>Thu, 12 Nov 2026 12:00:00 +0000</pubDate><source url="https://aistackcrunch.com/new/feed.xml">AI Stack Crunch</source><description>Google released Gemini 4 Hydrogen on Nov 12, a faster, lower-cost model for coding and agent work, with a 1M-token context window, priced at $1.50 per million input tokens and $7.50 per million output tokens. In Google Antigravity, you can now plan with Gemini 4 Carbon or Argon and have Hydrogen handle the coding and testing in the same conversation. (Source: Google)</description></item><item><title>Grok Bot on Mac can now work with Messages, Mail, Notes and more</title><link>https://x.com/bot/status/2120946968403095906</link><guid isPermaLink="false">2026-11-12|https://x.com/bot/status/2120946968403095906</guid><pubDate>Thu, 12 Nov 2026 12:00:00 +0000</pubDate><source url="https://aistackcrunch.com/new/feed.xml">AI Stack Crunch</source><description>On Nov 12, Grok Bot on Mac gained the ability to work with Messages, Mail, Notes, Reminders, Calendar, Location and Contacts. (Source: @bot)</description></item><item><title>Anthropic launches Claude Meetings in beta on Team and Enterprise plans</title><link>https://claude.com/resources/articles/claude-meetings</link><guid isPermaLink="false">2026-11-12|https://claude.com/resources/articles/claude-meetings</guid><pubDate>Thu, 12 Nov 2026 12:00:00 +0000</pubDate><source url="https://aistackcrunch.com/new/feed.xml">AI Stack Crunch</source><description>Anthropic launched Claude Meetings in beta, rolling out on Team and Enterprise plans. You add Claude to a Google Meet call as a visible participant; it takes notes and action items, answers questions in the meeting chat, and builds docs, whiteboards and other artifacts live that teammates can edit. Results land on a meeting page in claude.ai. Other video platforms are coming soon. Notes, transcripts and most answers are free during the beta, and in-call artifacts are free through Feb 10, 2027. It is not available during the beta for ZDR, HIPAA-ready or CMEK organizations. (Source: Anthropic)</description></item><item><title>Google launches the Gemini Decisions API for fast yes/no, choice and rubric decisions</title><link>https://blog.google/innovation-and-ai/technology/developers-tools/gemini-decisions-api/</link><guid isPermaLink="false">2026-11-11|https://blog.google/innovation-and-ai/technology/developers-tools/gemini-decisions-api/</guid><pubDate>Wed, 11 Nov 2026 12:00:00 +0000</pubDate><source url="https://aistackcrunch.com/new/feed.xml">AI Stack Crunch</source><description>Google introduced the Gemini Decisions API on Nov 11, powered by a new model, Gemini Decision 1 (gemini-decision-1), in the Gemini API via Google AI Studio. Instead of generating text, it scores a fixed set of options in one pass: yes/no checks, multiple choice (2 to 255 options) and ordered rubric scores, each with a 0 to 1 confidence score so code can send unclear cases to a fallback or a human. One request can carry up to 1,000 questions against a shared input of text, images, audio, video or PDFs (up to 256K tokens). Google quotes 150 to 250 ms end to end and $0.10 per million input tokens with no output tokens billed, plus a no-cost tier for prototyping. Typical uses are tool selection, model routing, policy checks and grading. Benchmark and latency claims are Google&#x27;s, not ours. (Source: Google)</description></item><item><title>AnimatedWebs and MotionVault: animated templates and a one-sentence site builder</title><link>https://animatedwebs.com/</link><guid isPermaLink="false">2026-11-11|https://animatedwebs.com/</guid><pubDate>Wed, 11 Nov 2026 12:00:00 +0000</pubDate><source url="https://aistackcrunch.com/new/feed.xml">AI Stack Crunch</source><description>AnimatedWebs is a library of 1,623 animated templates (888 full websites, plus sections, backgrounds and UI components) with 476 live demos and build prompts for Lovable, v0, Bolt, Claude Code and Cursor, plus an official MCP server and a Claude Code plugin. The same team also makes MotionVault, which turns one sentence about a business into an animated website, writes the copy and publishes it in one click. Free MotionVault sites go live on a shipaw.site link with a badge. Both are from Like a King Inc., the company behind the Reges Town AI city demo. (Source: AnimatedWebs)</description></item><item><title>Grok Bot 1.0 launches as shareable AI teammates for Slack, Teams and meetings</title><link>https://x.com/bot/status/2120582957782200586</link><guid isPermaLink="false">2026-11-11|https://x.com/bot/status/2120582957782200586</guid><pubDate>Wed, 11 Nov 2026 12:00:00 +0000</pubDate><source url="https://aistackcrunch.com/new/feed.xml">AI Stack Crunch</source><description>On Nov 11, the @bot account (affiliated with SpaceX Super Intelligence) introduced Grok Bot 1.0, describing it as AI teammates that work alongside your whole team. The official post says you can share Bots, collaborate with them in Slack and Teams, invite them to meetings, and watch them learn and become more helpful. (Source: @bot)</description></item><item><title>Graphed publishes UGC Studio, a HeyGen-style UGC ad skill for coding agents</title><link>https://github.com/graphed-inc/marketing-skills/tree/main/skills/ugc-studio</link><guid isPermaLink="false">2026-11-10|https://github.com/graphed-inc/marketing-skills/tree/main/skills/ugc-studio</guid><pubDate>Tue, 10 Nov 2026 12:00:00 +0000</pubDate><source url="https://aistackcrunch.com/new/feed.xml">AI Stack Crunch</source><description>Graphed&#x27;s public marketing-skills catalog (MIT license) includes UGC Studio, a skill a coding agent can install from Claude, Cursor or ChatGPT. Official docs say it deploys a HeyGen-style video studio into your Graphed project: write scripts from your website, pick avatars, and render a grid of vertical talking-head ads in parallel, each checked word by word against the script. &quot;Write from website&quot; pulls buyers&#x27; pain points from Reddit and writes 10 scripts. The skill bills in Graphed credits (1 credit = 1 cent) at about $4.80 per 30 seconds of video, and it tells the agent not to render until you approve the cost. The same catalog is listed at graphed.com/skills. The ugc-studio skill files appear in the public GitHub repo as of Nov 10, 2026. (Source: Graphed)</description></item><item><title>Google releases Gemini 4 Carbon, its strongest model, at $2/$10 per million tokens (intro price)</title><link>https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-4-carbon-now-available/</link><guid isPermaLink="false">2026-11-10|https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-4-carbon-now-available/</guid><pubDate>Tue, 10 Nov 2026 12:00:00 +0000</pubDate><source url="https://aistackcrunch.com/new/feed.xml">AI Stack Crunch</source><description>Google made Gemini 4 Carbon generally available on Nov 10. Google pitches it for the hardest work (long multi-step coding and debugging, scientific research, heavy document analysis) while Gemini 4 Argon stays the everyday pick, and says Carbon uses more tokens than Argon to reason and verify. API access is for paying developers in Google AI Studio and Gemini for business at $2 per million input and $10 per million output tokens until Nov 30, 2026, then $4 and $20. Consumers get it in the Gemini app and AI Mode in Search on Google AI Ultra. Google also previewed /swarm in Antigravity, which turns one prompt into a team of agents. Benchmark rankings cited are Google&#x27;s and its partners&#x27;, not ours. (Source: Google)</description></item><item><title>Claude adds Whiteboards and Sheets (beta) and renames Dashboards to Claude Data</title><link>https://claude.com/resources/articles/sheets-whiteboards-data</link><guid isPermaLink="false">2026-11-10|https://claude.com/resources/articles/sheets-whiteboards-data</guid><pubDate>Tue, 10 Nov 2026 12:00:00 +0000</pubDate><source url="https://aistackcrunch.com/new/feed.xml">AI Stack Crunch</source><description>Anthropic launched Claude Whiteboards (a shared board your team brainstorms on with Claude) and Claude Sheets (spreadsheets with tabs, formulas and drop-downs that Claude builds and edits with you) in beta, and made Claude Dashboards generally available as Claude Data, with certified queries and live sharing in beta. All are on paid plans, on web, desktop and mobile. Sheets is basic for now: export to .xlsx or CSV, or use Claude for Excel, when you need more. Boards can move into Lucid, Miro or Vani by Zoho, and sheets into Airtable, Asana, monday.com or Smartsheet. Claude Data needs a data connector such as a warehouse or CRM. (Source: Anthropic)</description></item><item><title>Cursor Projects is now multiplayer, in beta for all users</title><link>https://cursor.com/blog/projects-multiplayer</link><guid isPermaLink="false">2026-11-10|https://cursor.com/blog/projects-multiplayer</guid><pubDate>Tue, 10 Nov 2026 12:00:00 +0000</pubDate><source url="https://aistackcrunch.com/new/feed.xml">AI Stack Crunch</source><description>On Nov 10, Cursor said Projects, its coordinator-led fleet of agents for large work, is now multiplayer. You can invite teammates into a Project and plan, build and ship alongside everyone&#x27;s agents. Each person still has a one-to-one chat with their own coordinator and worker agents, and a shared context layer holds docs, diagrams and specs so a decision in one chat is available to every agent in the Project. A shared task board shows who is doing what, what agents are working on, and the PR behind each task. Teammates&#x27; agents can also coordinate with each other. It is rolling out in beta to all Cursor users; create a Project from the left-hand nav and invite teammates. (Source: Cursor)</description></item><item><title>Stripe says OpenRouter is now part of Stripe</title><link>https://x.com/stripe/status/2120240651778355306</link><guid isPermaLink="false">2026-11-10|https://x.com/stripe/status/2120240651778355306</guid><pubDate>Tue, 10 Nov 2026 12:00:00 +0000</pubDate><source url="https://aistackcrunch.com/new/feed.xml">AI Stack Crunch</source><description>On Nov 10, Stripe posted that OpenRouter is now part of Stripe, and that together they will help businesses explore new models and optimize token routing and usage. OpenRouter confirmed the same day that it is officially part of Stripe, that its commitment to neutrality stays the same, and that models and providers still compete on quality, reliability, and price. Stripe first announced on Aug 19, 2026 that it had agreed to acquire OpenRouter, a model gateway and routing platform used to route and optimize token usage; OpenRouter said it would keep the same name, product, and roadmap, and that the transaction was subject to customary closing conditions. Deal terms were not disclosed. No separate November newsroom close post was found; the Nov 10 official posts are the public close statements. (Source: Stripe)</description></item><item><title>Meta Muse Ember: a $0.10 per million token model for agent subtasks</title><link>https://research.meta.ai/blog/muse-ember</link><guid isPermaLink="false">2026-11-09|https://research.meta.ai/blog/muse-ember</guid><pubDate>Mon, 09 Nov 2026 12:00:00 +0000</pubDate><source url="https://aistackcrunch.com/new/feed.xml">AI Stack Crunch</source><description>Meta released Muse Ember (muse-ember-1.0), its smallest and cheapest Muse model, built for the repetitive work agents do all day: tool calls, coding subtasks, routing, extraction and long document parsing. It costs $0.10 per million input tokens and $0.50 per million output (prompts up to 200K tokens), with a 1M-token context window and up to 512K output tokens. Meta says it matches Claude Haiku 5.5 on the Artificial Analysis Index at a lower cost per task; benchmark figures are Meta&#x27;s and Artificial Analysis&#x27;s, not ours. Available on Muse Code and the Meta Model API, plus Google Cloud and Oracle Cloud. (Source: Meta AI)</description></item><item><title>Anthropic launches Claude Fable 5.2 and Mythos 5.2</title><link>https://www.anthropic.com/claude-fable-and-mythos-5-2</link><guid isPermaLink="false">2026-11-09|https://www.anthropic.com/claude-fable-and-mythos-5-2</guid><pubDate>Mon, 09 Nov 2026 12:00:00 +0000</pubDate><source url="https://aistackcrunch.com/new/feed.xml">AI Stack Crunch</source><description>Anthropic released Claude Fable 5.2 and Mythos 5.2 for ambitious open-ended technical work. Fable and Mythos are the same underlying model; Fable ships with cyber and biology safeguards for all users, while Mythos is for vetted organizations. API pricing is $8 per million input tokens and $40 per million output ($16/$80 in optional fast mode, up to 2.5x faster). Anthropic recommends Opus 5.5 as the daily driver and Fable for harder work. Availability: all platforms via claude-fable-5-2, with extra Fable usage for eligible paid plans this week. (Source: Anthropic)</description></item><item><title>Notion says a Fireworks AI breach exposed some Notion AI chat transcripts</title><link>https://www.notion.com/help/security-incident-involving-a-third-party-ai-vendor</link><guid isPermaLink="false">2026-11-08|https://www.notion.com/help/security-incident-involving-a-third-party-ai-vendor</guid><pubDate>Sun, 08 Nov 2026 12:00:00 +0000</pubDate><source url="https://aistackcrunch.com/new/feed.xml">AI Stack Crunch</source><description>Notion published a notice that a security incident at its inference vendor Fireworks AI may affect under 1% of Notion workspaces. Fireworks confirmed the affected dataset on Nov 8; it held Notion AI chat transcripts used for the Insights feature, without user or workspace labels and without tool call results. Notion has disconnected from Fireworks, disabled Insights and contacted affected workspace owners, who can review the data under Security incident review. If you pasted passwords or API keys into Notion AI, Notion recommends rotating them. (Source: Notion Help Center)</description></item><item><title>Qoder Qoduck: a free daily allowance for a stealth coding model</title><link>https://docs.qoder.com/events/qoduck-free</link><guid isPermaLink="false">2026-11-07|https://docs.qoder.com/events/qoduck-free</guid><pubDate>Sat, 07 Nov 2026 12:00:00 +0000</pubDate><source url="https://aistackcrunch.com/new/feed.xml">AI Stack Crunch</source><description>Qoder (the AI coding app from Alibaba Cloud) started a limited-time promotion on Nov 7: individual users, including Free accounts, get a daily free allowance for Qoduck, a stealth model whose identity Qoder has not disclosed. Inside the allowance calls cost no Credits, even at the highest thinking setting; after it, Qoduck bills at 0.5x Credits. Paid plans get more daily uses, but the exact count is not published, and Teams and Enterprise plans are excluded. It works in the main Qoder app only, not the IDE, JetBrains plugin or CLI, and the end date has not been announced. (Source: Qoder docs)</description></item><item><title>Thinking Machines Inkling 2 Small: a 594B MoE you can fine-tune on Tinker</title><link>https://thinkingmachines.ai/news/inkling-2-small/</link><guid isPermaLink="false">2026-11-05|https://thinkingmachines.ai/news/inkling-2-small/</guid><pubDate>Thu, 05 Nov 2026 12:00:00 +0000</pubDate><source url="https://aistackcrunch.com/new/feed.xml">AI Stack Crunch</source><description>Thinking Machines Lab released Inkling 2 Small on Nov 5, the first model of its second generation. It is a Mixture-of-Experts model with 594B total and 17B active parameters, takes text and image input, supports variable reasoning effort and a context window of up to 1M tokens, and has no audio support. The lab reports a score of 31 on the Artificial Analysis Intelligence Index, with the biggest gains in agentic coding and vision. It is available now on Tinker for inference, sampling and fine-tuning, plus the Tinker Playground. Weights are closed at launch; the lab says it intends to release them under Apache 2.0 after external safety testing, which has not happened yet. (Source: Thinking Machines Lab)</description></item><item><title>Kino AI: a code-first video editor your agents and teammates edit together</title><link>https://kino.ai</link><guid isPermaLink="false">2026-11-04|https://kino.ai</guid><pubDate>Wed, 04 Nov 2026 12:00:00 +0000</pubDate><source url="https://aistackcrunch.com/new/feed.xml">AI Stack Crunch</source><description>Kino (YC S23) relaunched on Nov 4 as a desktop video editor and media manager where Claude Code, Codex and your teammates work on the same timeline in real time. Every edit, motion graphic and comment is stored as code (built on Remotion), and footage is transcribed and indexed so agents can find clips. It handles pro formats like ProRes, BRAW and R3D and renders locally. The full editor is free, with MCP access for your own agent; Pro is $25 per editor per month for unlimited MCP calls, share links, more storage and team editing. Desktop app (Electron) only for now, no browser version yet. (Source: Kino (Launch HN))</description></item><item><title>Oumi FAiE builds AI agents and retrains the models behind them</title><link>https://oumi.ai/product-launch/fall26</link><guid isPermaLink="false">2026-11-04|https://oumi.ai/product-launch/fall26</guid><pubDate>Wed, 04 Nov 2026 12:00:00 +0000</pubDate><source url="https://aistackcrunch.com/new/feed.xml">AI Stack Crunch</source><description>Oumi&#x27;s Frontier AI Engineer (FAiE) can now build agents, not just custom models. You describe what an agent should do and FAiE helps define its tools and environment, design evaluations, curate data and train the model underneath, then turns production failures into new test and training data. Teams with agents elsewhere can keep their setup and connect it for evaluation and training (Oumi calls this Bring Your Own Harness). New training options include DPO, with GRPO reinforcement learning in private preview. Aimed at enterprise teams; FAiE asks for approval before any job that costs money. No public pricing in the launch. (Source: Oumi)</description></item><item><title>Fireworks Nexus Desktop: open models inside Codex, Claude Desktop and OpenCode</title><link>https://fireworks.ai/blog/fireworks-unveils-new-products-at-forge-to-help-companies-build-specialized-intelligence</link><guid isPermaLink="false">2026-11-03|https://fireworks.ai/blog/fireworks-unveils-new-products-at-forge-to-help-companies-build-specialized-intelligence</guid><pubDate>Tue, 03 Nov 2026 12:00:00 +0000</pubDate><source url="https://aistackcrunch.com/new/feed.xml">AI Stack Crunch</source><description>At its first Forge conference on Nov 3, Fireworks launched Nexus Desktop, which joins the Nexus CLI. It brings open models into the AI coding and work apps you already use (Codex, Claude Desktop and OpenCode) next to your existing subscriptions, picks a model per task, and gives teams one view of spend. Pro plans start at $8 per month and Teams at $10 per seat per month. Fireworks also put an ultra-fast inference tier into public preview. Speed figures are Fireworks&#x27; own benchmarks. (Source: Fireworks AI)</description></item><item><title>Prisma plugin for ChatGPT: build, add a database and deploy from one prompt</title><link>https://www.prisma.io/blog/one-chatgpt-plugin-to-deploy-it-all</link><guid isPermaLink="false">2026-11-03|https://www.prisma.io/blog/one-chatgpt-plugin-to-deploy-it-all</guid><pubDate>Tue, 03 Nov 2026 12:00:00 +0000</pubDate><source url="https://aistackcrunch.com/new/feed.xml">AI Stack Crunch</source><description>Prisma launched an official ChatGPT plugin on Nov 3 that connects ChatGPT and Codex to Prisma ORM, Prisma Postgres and Prisma Compute. You describe an app (Prisma&#x27;s example is a workout tracker) and ChatGPT writes the data layer, creates and connects the database, and deploys it, without switching dashboards or CLIs. Handy for solo builders who want a working app with a real database fast. Costs depend on your Prisma Postgres and Compute usage. (Source: Prisma)</description></item><item><title>Hume Audio Decision API: ask a recording a question, get probabilities back</title><link>https://www.hume.ai/blog/introducing-hume-audio-decision-api</link><guid isPermaLink="false">2026-11-02|https://www.hume.ai/blog/introducing-hume-audio-decision-api</guid><pubDate>Mon, 02 Nov 2026 12:00:00 +0000</pubDate><source url="https://aistackcrunch.com/new/feed.xml">AI Stack Crunch</source><description>Hume launched an audio-native decision model on Nov 2. You send an MP3 or WAV file (5 seconds to 5 minutes) with plain-language questions such as &quot;Does the caller sound frustrated?&quot; and get back probabilities for yes/no, multiple-choice, scale or time-span answers, with no transcript in between. Useful for reviewing support calls and testing voice agents. Hume reports 70.7% agreement with a human panel in its own evaluation. Pricing is $0.47 per million input tokens with a 1M-token trial credit, but API access is through a waitlist for now. (Source: Hume AI)</description></item><item><title>Momentic Mo is generally available: an AI QA engineer that tests your app end to end</title><link>https://momentic.ai/</link><guid isPermaLink="false">2026-11-02|https://momentic.ai/</guid><pubDate>Mon, 02 Nov 2026 12:00:00 +0000</pubDate><source url="https://aistackcrunch.com/new/feed.xml">AI Stack Crunch</source><description>Momentic (YC W24) took Mo out of beta and launched it on Hacker News on Nov 2. You describe a flow in plain English and Mo tests it against your web app or hosted iOS and Android devices, then reports bugs after a separate agent reproduces each one. It can run from a coding agent, a PR comment or the qa CLI and post a review on the pull request. Free to start with no credit card. Benchmarks on its site are Momentic&#x27;s own. (Source: Momentic)</description></item><item><title>Cursor security notice: malicious Claude Code extension versions 2.1.312 and 2.1.313</title><link>https://cursor.com/blog/claude-code-extension-security-notice</link><guid isPermaLink="false">2026-10-29|https://cursor.com/blog/claude-code-extension-security-notice</guid><pubDate>Thu, 29 Oct 2026 12:00:00 +0000</pubDate><source url="https://aistackcrunch.com/new/feed.xml">AI Stack Crunch</source><description>Cursor says maliciously modified versions 2.1.312 and 2.1.313 of the Anthropic.claude-code extension, not built by Anthropic, were served from the Cursor VS Code Extension Marketplace for Windows and macOS between about 2:27 AM and 1:35 PM PT on Oct 29. The code tried to install a persistent SSH backdoor and sent host details to an attacker server, which Cursor says is now offline. Cursor removed the versions, rebuilt the marketplace and rotated its keys. If you installed or updated the extension in that window, run Cursor&#x27;s one-line check on macOS: grep -qs svc-telemetry ~/.ssh/authorized_keys &amp;&amp; echo &quot;svc-telemetry key found&quot; || echo &quot;No key found; machine not impacted&quot;. A Windows PowerShell check and remediation scripts are in the notice; uninstalling alone may not be enough. (Source: Cursor)</description></item><item><title>Factory Droid 1.0: coding agent harness that runs tools from a script</title><link>https://factory.com/news/frontier-harness</link><guid isPermaLink="false">2026-10-29|https://factory.com/news/frontier-harness</guid><pubDate>Thu, 29 Oct 2026 12:00:00 +0000</pubDate><source url="https://aistackcrunch.com/new/feed.xml">AI Stack Crunch</source><description>Factory released Droid 1.0, the largest update yet to the harness behind its coding agent. The headline change is programmatic tool calling by default: the model writes a short sandboxed JavaScript program that can make many tool calls in one request, so raw files and logs stay out of its context and fewer model calls are needed. Factory reports it beats the native harnesses built for the same models on its own tests (vendor benchmarks, so test on your own work). Available now as a desktop app for macOS and Windows, a terminal CLI, and a TypeScript SDK; the launch post does not discuss pricing. (Source: Factory)</description></item><item><title>ElevenLabs SFX v3: sound effects from multi-part prompts</title><link>https://elevenlabs.io/blog/sound-effects-v3</link><guid isPermaLink="false">2026-10-29|https://elevenlabs.io/blog/sound-effects-v3</guid><pubDate>Thu, 29 Oct 2026 12:00:00 +0000</pubDate><source url="https://aistackcrunch.com/new/feed.xml">AI Stack Crunch</source><description>ElevenLabs launched SFX v3, its newest text-to-sound-effects model. It follows prompts that describe several sounds in one scene more reliably, handles music and vocals better, and is less likely to produce harsh artifacts on chimes, gongs and whistles than SFX v2. It is in ElevenCreative (Sound Effects tab, Flows, Studio, Music Editor), the mobile app, and the API as eleven_text_to_sound_v3. Note that the API docs list it as limited access that may not be available in every region, and v2 stays the API default. (Source: ElevenLabs)</description></item><item><title>Amp macOS Satellites: the coding agent gets a cloud Mac for iOS and Mac apps</title><link>https://ampcode.com/news/macos-satellites</link><guid isPermaLink="false">2026-10-29|https://ampcode.com/news/macos-satellites</guid><pubDate>Thu, 29 Oct 2026 12:00:00 +0000</pubDate><source url="https://aistackcrunch.com/new/feed.xml">AI Stack Crunch</source><description>Amp&#x27;s coding agent can now start a fresh cloud Mac with Xcode and simulators (ready in about 30 seconds) whenever a task needs macOS, while the main thread keeps running on its Linux machine. It builds, tests and runs iOS or Mac apps, sends back screenshots and logs, and you can watch or take over the Mac&#x27;s screen. Pricing is $2/hour (mac.medium) or $4/hour (mac.large), drawn from plan usage and then Amp credits. It is in beta behind a waitlist. (Source: Amp)</description></item><item><title>Pika H3 Flex: fast video with synced audio, open weights (with a big license catch)</title><link>https://pika.art/blog/pika-h3-flex</link><guid isPermaLink="false">2026-10-29|https://pika.art/blog/pika-h3-flex</guid><pubDate>Thu, 29 Oct 2026 12:00:00 +0000</pubDate><source url="https://aistackcrunch.com/new/feed.xml">AI Stack Crunch</source><description>Pika released H3 Flex, its post-trained version of MiniMax H3 that generates video with synchronized audio in six sampling steps and accepts up to 12 references (images, videos and audio). In Pika&#x27;s hosted product it finishes at up to 4K and can export separate audio stems. Pika also published the adapters and inference code, but local runs are 480p or 768p and need a large NVIDIA GPU (Pika&#x27;s tests used an H100 80GB). Read the license before you plan on self-hosting: the MiniMax H3 Community License excludes the United States, EU, UK and South Korea from its territory, so for most of our readers the hosted Pika product is the practical route. (Source: Pika)</description></item><item><title>Gemini 4 Carbon: Google&#x27;s new frontier model, cyber defenders first</title><link>https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-4-carbon/</link><guid isPermaLink="false">2026-10-28|https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-4-carbon/</guid><pubDate>Wed, 28 Oct 2026 12:00:00 +0000</pubDate><source url="https://aistackcrunch.com/new/feed.xml">AI Stack Crunch</source><description>Google DeepMind announced Gemini 4 Carbon, its strongest model so far, aimed at long-horizon coding and cybersecurity work. Google reports 88.2% on SEC-bench Pro and a score of 59 on the Artificial Analysis Intelligence Index v5. You cannot use it yet unless you are in the first cohort: access starts with a subset of cyber defenders and trusted partners, with developer, business and consumer access promised later. Introductory API price is $2 per million input tokens and $10 per million output tokens (rising to $4 and $20 after the intro period). (Source: Google)</description></item><item><title>Articul8 Bivia: a decision model for AI agents that can say &quot;not sure&quot;</title><link>https://articul8.ai/bivia</link><guid isPermaLink="false">2026-10-28|https://articul8.ai/bivia</guid><pubDate>Wed, 28 Oct 2026 12:00:00 +0000</pubDate><source url="https://aistackcrunch.com/new/feed.xml">AI Stack Crunch</source><description>Articul8, the Intel spinout, launched Bivia, a multimodal model that returns structured decisions (yes/no, choose, score, extract a field) with a confidence score instead of generated text, and points to the exact passage, image region or video timestamp behind each answer. It can abstain when evidence is thin. Available now by API with pay-per-input-token pricing and no subscription; Articul8 says the first 10,000 users get up to 1B free tokens. Accuracy claims so far come from Articul8&#x27;s own testing. (Source: Articul8)</description></item><item><title>Tasklet ships Tasklet MCP: delegate work and skills to 24/7 cloud agents from Claude, ChatGPT and Cursor</title><link>https://tasklet.ai/blog/2026-10-28-tasklet-mcp</link><guid isPermaLink="false">2026-10-28|https://tasklet.ai/blog/2026-10-28-tasklet-mcp</guid><pubDate>Wed, 28 Oct 2026 12:00:00 +0000</pubDate><source url="https://aistackcrunch.com/new/feed.xml">AI Stack Crunch</source><description>Tasklet (tasklet.ai), the USV, Lightspeed and YC-backed cloud agent OS for work ($20M raised at a $175M valuation), launched Tasklet MCP. You can now hand off tasks or copy Claude Code and Codex skills straight to your 24/7 Tasklet agents from inside Claude, ChatGPT, Cursor or Codex, while keeping shared team connections, cloud browser and per-tool approval cards in place. Free to try with 600 daily bonus credits; paid plans start at $25/mo with unlimited team seats. Now listed on Best AI Agent Builders and Best AI Assistants &amp; Bots. (Source: Tasklet)</description></item><item><title>AdLume (Lumy): Slack-first AI agent for Google and Meta Ads + 20 open-source Claude Skills</title><link>https://adlume.co/</link><guid isPermaLink="false">2026-10-28|https://adlume.co/</guid><pubDate>Wed, 28 Oct 2026 12:00:00 +0000</pubDate><source url="https://aistackcrunch.com/new/feed.xml">AI Stack Crunch</source><description>Founded by Marek Dabrowski, AdLume gives you Lumy, an AI media buyer that lives in Slack and connects to Google Ads, Meta Ads and Google Tag Manager. Instead of making black-box changes on its own, Lumy runs 400+ daily account and conversion-tracking checks and posts every proposed budget, keyword or creative move in Slack for your approval first ($10 one-time 7-day trial with no auto-renewal, then $89/mo). Alongside Lumy, Marek also open-sourced 20 Claude Skills for Paid Ads (plus 17 other marketing skill packs) free under the MIT license on GitHub. Now listed in Paid Ads on Best for Marketing &amp; Advertising. (Source: AdLume)</description></item><item><title>Vercel Labs agent-desktop 1.0: computer use for agents without pixel guessing</title><link>https://github.com/vercel-labs/agent-desktop</link><guid isPermaLink="false">2026-10-28|https://github.com/vercel-labs/agent-desktop</guid><pubDate>Wed, 28 Oct 2026 12:00:00 +0000</pubDate><source url="https://aistackcrunch.com/new/feed.xml">AI Stack Crunch</source><description>Vercel Labs shipped version 1.0 of agent-desktop, a free, open-source (Apache-2.0) CLI and C library that lets an AI agent read and operate desktop apps on macOS, Windows and Linux through the operating system&#x27;s accessibility tree. Each snapshot returns JSON with a stable reference for every clickable element, so the agent acts on a reference and checks again instead of guessing from screenshots. Install with npm install -g agent-desktop; macOS needs Accessibility permission. It is a developer building block for your own agent, not a ready-made assistant, and snapshots can contain private screen content, so keep them out of public logs. (Source: Vercel Labs)</description></item><item><title>Judgment Labs: agent monitoring and improvement platform opens to everyone</title><link>https://judgmentlabs.ai/</link><guid isPermaLink="false">2026-10-27|https://judgmentlabs.ai/</guid><pubDate>Tue, 27 Oct 2026 12:00:00 +0000</pubDate><source url="https://aistackcrunch.com/new/feed.xml">AI Stack Crunch</source><description>Judgment Labs opened general access to its &quot;continuous-improvement stack&quot; for AI agents. It reads production traces, groups recurring failures into Issues with a diagnosis, builds judges to monitor them, and lets you test a fix against real production cases before you ship. You can work with it from Slack or through its MCP server in Claude, Codex or Cursor. Backed by $32M led by Lightspeed. Worth a look if you run long-running agents in production; overkill for a single workflow. (Source: Judgment Labs)</description></item><item><title>Nativ: free app for running open AI models locally on a Mac</title><link>https://www.neywalabs.ai/products/nativ</link><guid isPermaLink="false">2026-10-27|https://www.neywalabs.ai/products/nativ</guid><pubDate>Tue, 27 Oct 2026 12:00:00 +0000</pubDate><source url="https://aistackcrunch.com/new/feed.xml">AI Stack Crunch</source><description>Neywa Labs, the new company from mlx-vlm author Prince Canuma, came out of stealth with Nativ, a macOS app for Apple Silicon that downloads and runs open models locally for chat, vision, code, audio and image generation, with live speed and memory stats. It also serves OpenAI- and Anthropic-compatible local APIs, so tools like Claude Code, Codex and Aider can point at it. No account or subscription; needs macOS 26 and enough memory for the model you pick. Good pick for private work on a well-specced Mac. (Source: Neywa Labs)</description></item><item><title>Instinct: the $10B text-and-call AI concierge taking on Muse and the Bot Wars</title><link>https://instinct.com/</link><guid isPermaLink="false">2026-10-27|https://instinct.com/</guid><pubDate>Tue, 27 Oct 2026 12:00:00 +0000</pubDate><source url="https://aistackcrunch.com/new/feed.xml">AI Stack Crunch</source><description>Founded by former Sierra researcher and Reflexion co-author Noah Shinn, Instinct (instinct.com) is an always-on personal agent with its own computer, phone number, and voice concierge that you text or call to book travel, cancel subscriptions, order groceries, and call businesses that lack online booking. After raising at a $2.5B valuation in August, it raised another $1B Series C at a $10B valuation from Sequoia, Benchmark, and Coatue, with over half of its nearly $1B annualized transaction volume coming from travel. Currently free in an invite-only beta (5 invites per user), and now listed on Best AI Assistants &amp; Bots and Best for Solopreneur Productivity. (Source: Instinct)</description></item><item><title>Agentell: agency accountability and client ROI layer for n8n and Make agents</title><link>https://agentell.ai/</link><guid isPermaLink="false">2026-10-27|https://agentell.ai/</guid><pubDate>Tue, 27 Oct 2026 12:00:00 +0000</pubDate><source url="https://aistackcrunch.com/new/feed.xml">AI Stack Crunch</source><description>An accountability platform built for agency managers running client AI agents and workflows on n8n and Make. Connects with a single API key (no SDK or developer instrumentation), watches every API dependency, re-runs failed steps, pauses runaway spend, and turns cost per qualified lead and cost per resolution into client-ready renewal reports. Currently onboarding founding agencies through its month-to-month Hands-On Help beta. (Source: Agentell)</description></item><item><title>HOL Guard: open-source security and policy layer for coding agents</title><link>https://hol.org/guard</link><guid isPermaLink="false">2026-10-27|https://hol.org/guard</guid><pubDate>Tue, 27 Oct 2026 12:00:00 +0000</pubDate><source url="https://aistackcrunch.com/new/feed.xml">AI Stack Crunch</source><description>An Apache-2.0 local policy layer that sits beside Claude Code, Cursor, Codex and MCP servers to check shell commands, secret reads and package installs before they run, prompting you when an action looks risky. Local protection is free forever (pipx install hol-guard &amp;&amp; hol-guard init); Guard Cloud starts at $4.99 a month ($15 Pro, $30 per seat Team) for synced policies and audit receipts. As HOL notes, a policy check is one review layer rather than a complete defense. (Source: HOL)</description></item><item><title>OpenAI GPT-6.1 Terra: cheaper, faster model built for computer use</title><link>https://openai.com/index/introducing-gpt-6-1-terra/</link><guid isPermaLink="false">2026-10-26|https://openai.com/index/introducing-gpt-6-1-terra/</guid><pubDate>Mon, 26 Oct 2026 12:00:00 +0000</pubDate><source url="https://aistackcrunch.com/new/feed.xml">AI Stack Crunch</source><description>OpenAI&#x27;s new mid-tier model focuses on visual reasoning, computer use (filling forms, working across desktop apps) and clearer writing. OpenAI says it matches GPT-6 Astra on its computer-use benchmark at one-tenth the price per token and runs over 3x faster. It is in ChatGPT Work and Codex for paid plans (not regular Chat) and in the API as gpt-6.1-terra at $1 per million input and $5 per million output tokens, with a 1M-token context. OpenAI itself still points the hardest coding work to Sol or Astra. (Source: OpenAI)</description></item><item><title>Entire: a Git forge built for agents and humans</title><link>https://entire.com/blog/the-git-forge-for-agent-human-collaboration</link><guid isPermaLink="false">2026-10-26|https://entire.com/blog/the-git-forge-for-agent-human-collaboration</guid><pubDate>Mon, 26 Oct 2026 12:00:00 +0000</pubDate><source url="https://aistackcrunch.com/new/feed.xml">AI Stack Crunch</source><description>Former GitHub CEO Thomas Dohmke&#x27;s Entire opened its Git forge for agent-human collaboration. Instead of classic pull requests it offers Trails, a review flow with risk, intent drift and slop monitors, plus distributed code hosting, CI integrations and continuous memory that keeps prompts and agent sessions with every commit. Entire says hosting is up to 89 times faster than competitors; that is the maker&#x27;s own benchmark, so try it on a side repo before moving anything important. (Source: Entire)</description></item><item><title>Strawberry 1.0: AI browser with built-in companions</title><link>https://strawberrybrowser.com/updates/strawberry-1-0</link><guid isPermaLink="false">2026-10-26|https://strawberrybrowser.com/updates/strawberry-1-0</guid><pubDate>Mon, 26 Oct 2026 12:00:00 +0000</pubDate><source url="https://aistackcrunch.com/new/feed.xml">AI Stack Crunch</source><description>Strawberry, a desktop browser with AI companions that research, draft follow-ups, pull data from sites and run scheduled tasks, hit version 1.0. It connects to Gmail, Slack and Notion, and browsing itself is free; credits are only spent when a companion helps. Plans start free (500 monthly credits), then $20, $85 and $250 per month. Good fit for small teams that live in the browser; check how far the free credits go first. (Source: Strawberry)</description></item><item><title>Moonshot Kimi K3.1 Pro: open-weight model with 1M-token context</title><link>https://www.kimi.com/en/blog/kimi-k3-1-pro</link><guid isPermaLink="false">2026-10-24|https://www.kimi.com/en/blog/kimi-k3-1-pro</guid><pubDate>Sat, 24 Oct 2026 12:00:00 +0000</pubDate><source url="https://aistackcrunch.com/new/feed.xml">AI Stack Crunch</source><description>Moonshot&#x27;s new top model, with a faster Highspeed version (2 to 3 times quicker). It is aimed at coding, 3D creation and financial research, and Moonshot reports 65.7% on Terminal-Bench 4.0, close to Claude Opus 5.5. A new 3D Make feature on kimi.com lets you preview, edit and download 3D models. Available now on Kimi.com, Kimi Work, Kimi Code and the Kimi API; full open weights are promised by Nov 4. Benchmarks are the maker&#x27;s own, so test it on your real work first. (Source: Moonshot AI)</description></item><item><title>Bot Wars: Grok Bot, Muse, Dots and Claude Duet</title><link>https://aistackcrunch.beehiiv.com/p/the-crunch-001-bot-wars</link><guid isPermaLink="false">2026-10-23|https://aistackcrunch.beehiiv.com/p/the-crunch-001-bot-wars</guid><pubDate>Fri, 23 Oct 2026 12:00:00 +0000</pubDate><source url="https://aistackcrunch.com/new/feed.xml">AI Stack Crunch</source><description>Four always-on AI agents with their own cloud computers launched within about 10 weeks: xAI&#x27;s Grok Bot (Aug 11, beta for SuperGrok and Cursor plans, built to run several bots as a team), Meta&#x27;s Muse (Sep 8, free with a usage limit in the US, also on WhatsApp), OpenAI&#x27;s Dots (late Sep, included with ChatGPT Pro and Business Premium) and Anthropic&#x27;s Claude Duet (Oct 21, US Claude Pro and Max waitlist). Each keeps working after you close the app and asks before it sends, pays or deletes. Muse is the only free starting option; Grok Bot is the one built for multi-agent teams; Claude Duet is aimed at professionals and small businesses once you&#x27;re off the waitlist. Plans and regions are moving fast, so check the maker&#x27;s page before you sign up. (Source: AI Stack Crunch)</description></item><item><title>Canva Craft: Canva&#x27;s first in-house image model</title><link>https://www.canva.com/newsroom/news/canva-craft/</link><guid isPermaLink="false">2026-10-22|https://www.canva.com/newsroom/news/canva-craft/</guid><pubDate>Thu, 22 Oct 2026 12:00:00 +0000</pubDate><source url="https://aistackcrunch.com/new/feed.xml">AI Stack Crunch</source><description>An image model built by Canva Research for design work. It plans the layout and any text first, then renders up to 2K in 30+ aspect ratios, so posters, infographics and product shots come out with legible text and elements you can edit one at a time. Available to all Canva users in Dream Lab and to developers through Canva&#x27;s public API (generations use AI credits); Canva MCP support is coming. Leaderboard rankings cited are from DesignArena. (Source: Canva)</description></item><item><title>Anthropic Claude Duet: a proactive assistant app (research preview)</title><link>https://claude.com/resources/articles/claude-duet</link><guid isPermaLink="false">2026-10-21|https://claude.com/resources/articles/claude-duet</guid><pubDate>Wed, 21 Oct 2026 12:00:00 +0000</pubDate><source url="https://aistackcrunch.com/new/feed.xml">AI Stack Crunch</source><description>A new Claude app that sends a morning brief from your connected apps, watches them through the day and messages you only when something needs you. You can also hand it tasks, from scheduling meetings to building dashboards, forecasts or web apps; it keeps working when you step away and checks with you before acting on your behalf. Built on Claude Opus 5.5, on iOS, Android and claude.ai. Research preview by waitlist, starting with US Pro and Max subscribers, using your plan&#x27;s existing usage limits. (Source: Anthropic)</description></item><item><title>Cursor Agent SDK: build and deploy agents as code</title><link>https://cursor.com/blog/agent-sdk</link><guid isPermaLink="false">2026-10-20|https://cursor.com/blog/agent-sdk</guid><pubDate>Tue, 20 Oct 2026 12:00:00 +0000</pubDate><source url="https://aistackcrunch.com/new/feed.xml">AI Stack Crunch</source><description>Cursor released the Agent SDK, the framework behind its own Bugbot and Security Reviewer agents. An agent is a project folder: instructions.md is the prompt, each file in tools/ is a typed action, and you test it with evals before deploying to Cursor or your own infrastructure with one command. Built for software teams automating code review, compliance or incident work, not a no-code builder. (Source: Cursor)</description></item><item><title>Mistral Voxtral 3: speech models + a voice agent SDK</title><link>https://mistral.ai/news/voxtral-3</link><guid isPermaLink="false">2026-10-20|https://mistral.ai/news/voxtral-3</guid><pubDate>Tue, 20 Oct 2026 12:00:00 +0000</pubDate><source url="https://aistackcrunch.com/new/feed.xml">AI Stack Crunch</source><description>New speech-to-text (Voxtral Transcribe 3, 50+ languages, diarization, end-of-turn detection) and text-to-speech (Voxtral Mini TTS 3, 35+ languages and 50+ dialects) plus a voice agent SDK built on LiveKit that handles listening, tool calls and speech in one stream, over SIP or WebRTC. Mistral&#x27;s API (EU, US and global endpoints) or self-hosted. Transcription from $0.0035/min, TTS $13 per 1M characters, voice agent API from $0.04/min. Quality comparisons are Mistral&#x27;s own evals. (Source: Mistral AI)</description></item><item><title>ElevenLabs Eleven Realtime v1 Mini and Max: speech-to-speech voice agents</title><link>https://elevenlabs.io/blog/realtime</link><guid isPermaLink="false">2026-10-20|https://elevenlabs.io/blog/realtime</guid><pubDate>Tue, 20 Oct 2026 12:00:00 +0000</pubDate><source url="https://aistackcrunch.com/new/feed.xml">AI Stack Crunch</source><description>Two speech-to-speech models that handle recognition, turn-taking, reasoning and voice as one system: v1 Mini for low-cost natural conversation, v1 Max for support, lead qualification and tool-heavy tasks (87.1% tool-task success vs 60.3% for Mini, per Artificial Analysis tests cited by ElevenLabs). 90+ languages and the full voice library. Use it in ElevenAgents or through a Realtime API compatible with OpenAI&#x27;s, so existing OpenAI Realtime code mostly needs a new endpoint and model name. Rolling out gradually to workspaces. (Source: ElevenLabs)</description></item><item><title>Perplexity Computer can now run fully local on a Mac</title><link>https://www.perplexity.ai/hub/blog/perplexity-computer-now-runs-fully-local-on-mac</link><guid isPermaLink="false">2026-10-20|https://www.perplexity.ai/hub/blog/perplexity-computer-now-runs-fully-local-on-mac</guid><pubDate>Tue, 20 Oct 2026 12:00:00 +0000</pubDate><source url="https://aistackcrunch.com/new/feed.xml">AI Stack Crunch</source><description>Perplexity&#x27;s Computer agent can now plan and run a whole task on an Apple silicon Mac, with the model (Qwen 3.6 or PPLX Pro, a post-trained Qwen) and the agent both on device. With permission it works across local files and apps like Notes, Messages, Mail and Numbers, and you can let it call the cloud for web research or connected apps. Perplexity recommends 36 GB of unified memory for fully local mode; Hybrid mode (cloud plus local) is still available. (Source: Perplexity)</description></item><item><title>Google Gemini API CLI (gemini-api), experimental preview</title><link>https://aistudio.google.com/learn/introducing-the-gemini-api-cli</link><guid isPermaLink="false">2026-10-19|https://aistudio.google.com/learn/introducing-the-gemini-api-cli</guid><pubDate>Mon, 19 Oct 2026 12:00:00 +0000</pubDate><source url="https://aistackcrunch.com/new/feed.xml">AI Stack Crunch</source><description>A single zero-dependency command-line binary for macOS, Linux and Windows that gives you, or any coding agent with shell access such as Claude Code, Codex or Cursor, direct access to the Gemini API: text, image, video, music and speech generation, document and YouTube analysis, transcription and Managed Agents like Deep Research. Media is written straight to disk with only the file path printed, and commands describe themselves for agents. Needs a Gemini API key. Separate from the existing Gemini CLI coding agent. (Source: Google)</description></item><item><title>Inception Mercury 3: a diffusion LLM at 1,000+ tokens/sec</title><link>https://www.inceptionlabs.ai/blog/introducing-mercury-3</link><guid isPermaLink="false">2026-10-19|https://www.inceptionlabs.ai/blog/introducing-mercury-3</guid><pubDate>Mon, 19 Oct 2026 12:00:00 +0000</pubDate><source url="https://aistackcrunch.com/new/feed.xml">AI Stack Crunch</source><description>Inception&#x27;s new diffusion model writes tokens in parallel: 1,068 tokens/sec on NVIDIA GPUs, a 1M-token context and OpenAI-compatible API. Inception says it matches or beats GPT-6 Luna and Claude Haiku 5.5 (vendor benchmarks). $0.20 / $0.70 per million input / output tokens, half price at launch; also on OpenRouter, Baseten and Vercel. Good fit for voice bots, support and browser agents where speed matters. (Source: Inception)</description></item><item><title>Notion Agents rebuilt: own browser, 24-hour tasks, cheaper Auto router</title><link>https://www.notion.com/releases/2026-10-19</link><guid isPermaLink="false">2026-10-19|https://www.notion.com/releases/2026-10-19</guid><pubDate>Mon, 19 Oct 2026 12:00:00 +0000</pubDate><source url="https://aistackcrunch.com/new/feed.xml">AI Stack Crunch</source><description>Notion Agents now get their own computer and browser, a &quot;Review for me&quot; mode with safety checks, and can run tasks for up to 24 hours. The new Auto model router claims results similar to top models at about 90% lower cost. You can connect a ChatGPT plan to your personal agent (BYOK is in private alpha for Enterprise), Notion MCP was overhauled, and AI usage moves to a new credit system. (Source: Notion)</description></item><item><title>Viggle JST-2: 3D scenes, characters and motion from one model</title><link>https://viggle.ai/jst2</link><guid isPermaLink="false">2026-10-19|https://viggle.ai/jst2</guid><pubDate>Mon, 19 Oct 2026 12:00:00 +0000</pubDate><source url="https://aistackcrunch.com/new/feed.xml">AI Stack Crunch</source><description>Viggle&#x27;s new world model generates explorable 3D scenes in about a second, animation-ready characters from a single image (no separate rigging), and full-body motion from text or video. Free to try in Viggle&#x27;s PINOC browser app, with API access for developers. Benchmark wins so far are Viggle&#x27;s own tests, and scenes are static for now. (Source: Viggle)</description></item><item><title>DeepSeek V4.1-Flash-1017 + Search &amp; Fetch APIs</title><link>https://api-docs.deepseek.com/news/news261017</link><guid isPermaLink="false">2026-10-17|https://api-docs.deepseek.com/news/news261017</guid><pubDate>Sat, 17 Oct 2026 12:00:00 +0000</pubDate><source url="https://aistackcrunch.com/new/feed.xml">AI Stack Crunch</source><description>Refreshed V4.1 Flash snapshot with better benchmarks, faster task completion and shorter outputs; live in DeepSeek&#x27;s app and API as deepseek-flash, with open weights on Hugging Face. DeepSeek also opened Search and Fetch APIs for agents and RAG at $0.75 per 1,000 calls each (free via server-side tools in DeepSeek Harness, Claude Code and Codex until billing starts Oct 31). (Source: DeepSeek)</description></item><item><title>ChatGPT Slides: build and edit decks in ChatGPT</title><link>https://help.openai.com/en/articles/20001605-creating-and-collaborating-on-slides-in-chatgpt</link><guid isPermaLink="false">2026-10-17|https://help.openai.com/en/articles/20001605-creating-and-collaborating-on-slides-in-chatgpt</guid><pubDate>Sat, 17 Oct 2026 12:00:00 +0000</pubDate><source url="https://aistackcrunch.com/new/feed.xml">AI Stack Crunch</source><description>Create a presentation from a chat, files, an outline or a template, edit slides directly or ask ChatGPT to change them, and co-edit with teammates in real time. Present in ChatGPT or export to PowerPoint, PDF or Google Slides. Rolling out to Pro, Business and Enterprise users with Space access. (Source: OpenAI)</description></item><item><title>Reflection releases Beam, its first open-weight model</title><link>https://reflection.ai/blog/beam-is-available</link><guid isPermaLink="false">2026-10-16|https://reflection.ai/blog/beam-is-available</guid><pubDate>Fri, 16 Oct 2026 12:00:00 +0000</pubDate><source url="https://aistackcrunch.com/new/feed.xml">AI Stack Crunch</source><description>Sparse MoE model (501B total, 23B active) built for coding, reasoning and long-horizon agents. Weights on Hugging Face under Apache 2.0, plus an open-source Mirror CLI harness. Also on Reflection&#x27;s API (public beta, free to start) and partners like OpenRouter, Together, Fireworks and Ollama. (Source: Reflection)</description></item><item><title>Meta releases Muse Spark Open weights</title><link>https://research.meta.ai/blog/muse-spark-open</link><guid isPermaLink="false">2026-10-15|https://research.meta.ai/blog/muse-spark-open</guid><pubDate>Thu, 15 Oct 2026 12:00:00 +0000</pubDate><source url="https://aistackcrunch.com/new/feed.xml">AI Stack Crunch</source><description>Open-weight, natively multimodal model (816B total, 42B active, 1M-token context) built for coding and agentic work. Weights on Hugging Face and major clouds; hosted via OpenRouter, Together, Groq, Fireworks and others. Runs on two M5 Ultra Mac Studios with MLX. Muse Model License has limited commercial terms. (Source: Meta AI)</description></item><item><title>Krea 3 image editing model</title><link>https://www.krea.ai/blog/introducing-krea-3</link><guid isPermaLink="false">2026-10-15|https://www.krea.ai/blog/introducing-krea-3</guid><pubDate>Thu, 15 Oct 2026 12:00:00 +0000</pubDate><source url="https://aistackcrunch.com/new/feed.xml">AI Stack Crunch</source><description>Krea&#x27;s new image editing model: most edits in under 8 seconds at native 2K, about $0.06 per 2K image, with consistent products and characters across edits. In the Krea app and API, and the default Krea image model as of Oct 16. (Source: Krea)</description></item><item><title>Gmail AI Inbox expands to paid Workspace</title><link>https://workspaceupdates.googleblog.com/2026/10/gmails-ai-inbox-is-rolling-out-to-paid-Workspace-users.html</link><guid isPermaLink="false">2026-10-15|https://workspaceupdates.googleblog.com/2026/10/gmails-ai-inbox-is-rolling-out-to-paid-Workspace-users.html</guid><pubDate>Thu, 15 Oct 2026 12:00:00 +0000</pubDate><source url="https://aistackcrunch.com/new/feed.xml">AI Stack Crunch</source><description>AI Inbox, which pulls suggested to-dos and catch-up summaries out of your email, is rolling out to paid Google Workspace Business, Enterprise and Frontline Plus users. US and EN-US only for now; gradual rollout over the next few weeks. (Source: Google Workspace)</description></item><item><title>Builder.io Agent-Native beta</title><link>https://www.builder.io/blog/introducing-agent-native</link><guid isPermaLink="false">2026-10-14|https://www.builder.io/blog/introducing-agent-native</guid><pubDate>Wed, 14 Oct 2026 12:00:00 +0000</pubDate><source url="https://aistackcrunch.com/new/feed.xml">AI Stack Crunch</source><description>Open-source TypeScript framework for agentic apps where the agent and UI share the same actions, data, and state. Works with MCP, A2A, Slack, and bring-your-own models; build locally or on Builder.io. (Source: Builder.io)</description></item><item><title>Vani Code Agent for VS Code</title><link>https://vani.ai/news/vani-code-agent-for-vs-code</link><guid isPermaLink="false">2026-10-14|https://vani.ai/news/vani-code-agent-for-vs-code</guid><pubDate>Wed, 14 Oct 2026 12:00:00 +0000</pubDate><source url="https://aistackcrunch.com/new/feed.xml">AI Stack Crunch</source><description>Free VS Code extension: a coding agent that edits files with diff review and runs commands only with your approval. Extension is free; API usage is billed through your Vani key. (Source: Vani AI)</description></item><item><title>Liquid AI LFM2.5-24B-A2B for on-device agents</title><link>https://www.liquid.ai/blog/lfm2-5-24b-a2b</link><guid isPermaLink="false">2026-10-13|https://www.liquid.ai/blog/lfm2-5-24b-a2b</guid><pubDate>Tue, 13 Oct 2026 12:00:00 +0000</pubDate><source url="https://aistackcrunch.com/new/feed.xml">AI Stack Crunch</source><description>Open-weight 24B MoE reasoning model with 2B active parameters and 131K context, post-trained for tool calling. Aimed at fast private agents on your own hardware (llama.cpp, MLX, vLLM, etc.). (Source: Liquid AI)</description></item><item><title>Klaviyo Composer marketing agent goes live</title><link>https://www.klaviyo.com/blog/composer-marketing-agent</link><guid isPermaLink="false">2026-10-13|https://www.klaviyo.com/blog/composer-marketing-agent</guid><pubDate>Tue, 13 Oct 2026 12:00:00 +0000</pubDate><source url="https://aistackcrunch.com/new/feed.xml">AI Stack Crunch</source><description>Klaviyo&#x27;s AI marketing agent is now available to every Klaviyo customer. It answers with cited evidence from your data, runs scheduled cloud tasks, and applies team Skills for recurring marketing work. (Source: Klaviyo)</description></item><item><title>Google Play: private vibe-coded Android apps</title><link>https://blog.google/products-and-platforms/platforms/google-play/google-play-updates-october-2026/</link><guid isPermaLink="false">2026-10-13|https://blog.google/products-and-platforms/platforms/google-play/google-play-updates-october-2026/</guid><pubDate>Tue, 13 Oct 2026 12:00:00 +0000</pubDate><source url="https://aistackcrunch.com/new/feed.xml">AI Stack Crunch</source><description>Build an Android app in Google AI Studio and launch it privately on your own phone via Play with a standard Google account. Rolling out in the U.S. for users 18+; no developer account required. (Source: Google Blog)</description></item><item><title>Gemini 4 Argon lands in GitHub Copilot</title><link>https://github.blog/changelog/2026-10-13-gemini-4-argon-in-github-copilot/</link><guid isPermaLink="false">2026-10-13|https://github.blog/changelog/2026-10-13-gemini-4-argon-in-github-copilot/</guid><pubDate>Tue, 13 Oct 2026 12:00:00 +0000</pubDate><source url="https://aistackcrunch.com/new/feed.xml">AI Stack Crunch</source><description>Google&#x27;s newest frontier model is generally available in Copilot for Pro+, Max, Business and Enterprise across VS Code, Copilot CLI, the coding agent and other IDEs. (Source: GitHub Changelog)</description></item><item><title>Gemini 4 Argon goes GA</title><link>https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-4-argon/</link><guid isPermaLink="false">2026-10-13|https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-4-argon/</guid><pubDate>Tue, 13 Oct 2026 12:00:00 +0000</pubDate><source url="https://aistackcrunch.com/new/feed.xml">AI Stack Crunch</source><description>Google&#x27;s new top model for coding, science and agentic work, 1M token limit, intro price $2/$10 per M tokens until Nov 30. (Source: Google Blog)</description></item><item><title>Meta launches Muse Wonder &amp; Muse Wonder Full</title><link>https://research.meta.ai/blog/muse-wonder</link><guid isPermaLink="false">2026-10-12|https://research.meta.ai/blog/muse-wonder</guid><pubDate>Mon, 12 Oct 2026 12:00:00 +0000</pubDate><source url="https://aistackcrunch.com/new/feed.xml">AI Stack Crunch</source><description>Meta&#x27;s most capable agent models, with Active Memory and Active Focus; available via Meta Model API and Muse Code. (Source: Meta AI)</description></item><item><title>Synthesia Syren (beta): video from a single prompt</title><link>https://www.synthesia.io/post/syren-studio</link><guid isPermaLink="false">2026-10-09|https://www.synthesia.io/post/syren-studio</guid><pubDate>Fri, 09 Oct 2026 12:00:00 +0000</pubDate><source url="https://aistackcrunch.com/new/feed.xml">AI Stack Crunch</source><description>Agent builds full videos with motion graphics, avatars, voiceover and b-roll, then refines in chat. (Source: Synthesia)</description></item><item><title>Garuda: open-source coding agent with an OS sandbox</title><link>https://madhu-sv.github.io/garuda/blog/introducing-garuda/</link><guid isPermaLink="false">2026-10-08|https://madhu-sv.github.io/garuda/blog/introducing-garuda/</guid><pubDate>Thu, 08 Oct 2026 12:00:00 +0000</pubDate><source url="https://aistackcrunch.com/new/feed.xml">AI Stack Crunch</source><description>Apache-2.0 agent for terminal, VS Code and Zed. Commands run in Seatbelt/bubblewrap; edits show diffs you approve; audit log included. Works with Claude, OpenRouter and local models. (Source: Garuda)</description></item><item><title>Cursor CLI adds Claude Opus 5.5 &amp; Sonnet 5.5 on Bedrock</title><link>https://cursor.com/docs/cli/changelog</link><guid isPermaLink="false">2026-10-08|https://cursor.com/docs/cli/changelog</guid><pubDate>Thu, 08 Oct 2026 12:00:00 +0000</pubDate><source url="https://aistackcrunch.com/new/feed.xml">AI Stack Crunch</source><description>Bring-your-own-key Bedrock users can now pick the newest Claude models in Cursor&#x27;s CLI agent. (Source: Cursor changelog)</description></item><item><title>Claude Code v2.1.295</title><link>https://github.com/anthropics/claude-code/releases/tag/v2.1.295</link><guid isPermaLink="false">2026-10-08|https://github.com/anthropics/claude-code/releases/tag/v2.1.295</guid><pubDate>Thu, 08 Oct 2026 12:00:00 +0000</pubDate><source url="https://aistackcrunch.com/new/feed.xml">AI Stack Crunch</source><description>Blocking hooks on failure, terminal status protocol, VS Code file rows and many fixes. (Source: GitHub)</description></item><item><title>GPT-6 with Intelligent UI rolls out to ChatGPT</title><link>https://openai.com/index/gpt-6-for-everyone/</link><guid isPermaLink="false">2026-10-07|https://openai.com/index/gpt-6-for-everyone/</guid><pubDate>Wed, 07 Oct 2026 12:00:00 +0000</pubDate><source url="https://aistackcrunch.com/new/feed.xml">AI Stack Crunch</source><description>ChatGPT answers with interactive UIs: charts, forms, buttons and on-the-fly mini tools. Free tier next day. (Source: OpenAI)</description></item><item><title>Cursor ships iOS app beta + faster Bugbot</title><link>https://trendintech.com/cursor-ai-code-editor-ios-app-bugbot-2026/</link><guid isPermaLink="false">2026-10-06|https://trendintech.com/cursor-ai-code-editor-ios-app-bugbot-2026/</guid><pubDate>Tue, 06 Oct 2026 12:00:00 +0000</pubDate><source url="https://aistackcrunch.com/new/feed.xml">AI Stack Crunch</source><description>Launch and steer cloud coding agents from your phone, plus hosted-repo beta. (Source: TrendInTech)</description></item></channel></rss>
