Archive, page 6
An Apache 2.0 benchmark where a model reads one unstructured state and answers five typed questions, each as a probability distribution. Schema follows TypeSafe AI primitives so it runs against any matching API.
Twelve separate frozen data configs from the Open-Jev project, turning a state and question into a yes/no probability, choice distribution, label probabilities or numeric decision. Includes source code and manifests.
A Parquet dataset of 64,964 normalized Indian public information document texts with source titles, URLs, extraction quality and topic labels, generated October 2026.
An MIT-licensed dataset of over 5,500 RL tasks in code and data science for hill-climbing small models, with 144 held-out tasks. A 2B model was trained on the tasks, with a curriculum run compared to shuffled.
GitHub Copilot for JetBrains IDEs gets new controls for choosing models and MCP servers at startup, plus chat improvements, available to users now.
MORENA is a 1.5 billion parameter open-weight language model trained from scratch for 12 African languages. Developers can download the weights and run it themselves.
Agibot is a personal AI agent app for Mac that runs always on, letting you delegate tasks to it. It is free forever, according to its Product Hunt listing.
ONLYOFFICE 10.0, an open-source office suite for documents, spreadsheets and presentations, adds AI tools, modular apps and stronger security.
Talorys is an open-source personal AI agent you can self-host on Cloudflare's free tier, so you can run your own assistant without paying for servers. It drew 182 points on Hacker News.
Farol is a macOS terminal app that tracks multiple coding agents and flags which one needs your attention, so you do not have to check each session manually.
Anthropic's Claude is now available inside Google Docs, Sheets and Slides, Tom's Guide reports, with seven suggested uses for the new integration.
Claude Code Projects is now rolling out to all Pro and Max subscribers who were on the waitlist, according to Pasquale Pillitteri.
ChatGPT subscribers can now use their existing plan to pay for usage of Devin, the AI software engineering agent, without a separate purchase, Startup Fortune reports.
Nace AI has open-sourced Drex 1.5, a 9B model that returns a probability for every option in one forward pass. It scores 58.08 on Decision Index 0.3.1 and reads up to 128K tokens.
Lune is a search engine designed for scientific AI agents, letting them look up research material. It launched on Product Hunt.
Google's Gemini app now gives free-plan users only an Auto model option, removing manual model selection. Everyone gets access to thinking or effort levels, rolling out now.
Maildun is a Mac app for designing on-brand emails by hand, with AI assistance, or through MCP. It was launched on Product Hunt.
Claude Managed Agents now only fetch URLs that already appeared in the session, limiting arbitrary web fetches. Developers using the agent service see the change.
PixRater is a photo tool for culling and rating images faster, launched on Product Hunt.
AgentDock lets developers answer Claude Code and Codex prompts from a single inbox in any terminal. Launched on Product Hunt.
pydantic-ai, a Python framework for building AI agents, shipped releases 2.52 to 2.55. Version 2.55 requires Python 3.11 or newer; older Python 3.10 installs resolve to 2.54.
OpenAI's Decisions API, now in public beta on GPT-6 Luna, returns typed probabilities, choices and scores about 10x faster than the Responses API. Billed at $0.10 per 1M input tokens, with no output charges.
CIQ's Fuzzball 4.3 runs validated open-weight models on a customer's own GPUs with one command. Targets self-hosted AI deployment.
Version 0.162.1 of Codex CLI, OpenAI's local coding agent, installable via npm i -g @openai/codex.
OpenAI rolled out composer predictions in Codex as a beta for eligible Pro users, suggesting input completions while typing prompts.
Microsoft released Microsoft-Decision-1, a decision-scoring model for fast decision-making, built on Alibaba's open-weight Qwen3.5-9B. It is available on Hugging Face and Microsoft Foundry.
FloorGen is a free web tool that creates floor plans from a hand-drawn sketch or a text prompt. It is listed on Product Hunt.
Grok Bot now has its own email address, letting users claim one, according to 9to5Mac. It is the second significant upgrade to the bot this week.
Review of GMKtec's EVO-X3, a Ryzen AI Max+ 395 mini-PC with 128GB memory and an OCuLink port, positioned as a local AI workstation.
KernelAI is a private, offline AI app for iPhone and Android that runs models on the device without sending data to servers.
robot-heads is an open-source React component that renders animated 3D robot heads with LED faces, showing what an AI agent is doing. Written in TypeScript; 165 stars in first week.
Google's AI watch face generator, launched last year, is rolling out to Pixel Watch 4 models, 9to5Google reports, alongside other updates.
GitGlow is a developer tool for reviewing your own code and the changes your AI coding agents make before you ship them. Launched on Product Hunt.
Artificial Analysis scores Mistral's new Large 4 at 38, behind open models from Xiaomi, Z.ai, Moonshot and DeepSeek. A third-party guide lists API setup at $4.18 per million tokens.
PocketWebTools for Mac bundles local AI chat, transcription and video tools in one app that runs on your Mac. Launched on Product Hunt.
Claude can now turn data into live dashboards and animations. Claude Docs, Slides and Design are also out of beta and available to free users.
OpenPilot is an open-source desktop AI agent that can run with whichever model you choose. Launched on Product Hunt.
ONNX Runtime, a library for running ML models across hardware, released 1.31.0, stabilizing model-package and EPContext APIs and improving CPU model loading and quantized MoE inference.
Qwen-Image-2.1-Turbo is a text-to-image model that generates images in 8 steps, with 7B parameters, on Hugging Face. An uncensored GGUF version has also been posted, with 426 likes and 2,138 downloads.
iwant is a command-line tool that finds available GPUs and hosts open models on them in one command. Launched on Product Hunt.