WIRED reports that Rabbit is launching OS3, a cross-platform agent meant to run on screens people already use, two years after the company pursued dedicated AI hardware to bypass mobile apps.
Each batch appears in the Batches tab of OpenRouter logs with its model, provider, status, and cost. Inputs and results are kept for 30 days or until the batch is deleted.
Images and files must be public URLs, not base64. Audio and video input are not accepted, and OpenRouter’s own web plugin is not available in batch.
OpenRouter’s Batch API is available for work that does not need an immediate answer. It cuts per-token pricing in half on most models and returns results within 24 hours.
The desk device shows idle, working, needs-a-decision, and done. Wahlin used Copilot CLI to create several animations and a web-based character studio, and published the project as esp32-agent-companion.
Free missions, each under an hour, take anyone from a blank page to a working app and add the project to a Handshake profile, with no installation or setup.
Bloomberg reports that the Pentagon is expanding Palantir’s Maven AI system to help vet military targets and adding safeguards to reduce civilian harm.
LLM requests can be classified automatically with typesafeai's Jev in OpenRouter Classifiers, and the results can be analyzed in Explore.
PayPal says it is partnering with Meta so PayPal customers can shop and check out with their Muse personal AI agents at PayPal merchants worldwide.
Arena.ai gives Qwen-Image-2.1 1,228 points and first place among open models, 17th overall, four points behind Gemini-3-pro-image-preview and 11 behind GPT-Image-1.5-high-fidelity.
Sam Altman says people outside AI labs should have a real say in how AI develops and a clear way to judge whether it is safe. He says standards should keep power from concentrating so new and open-model companies can compete, help countries and companies learn from failures, and be led by the United States.
An Origin repository can be connected to a Cloudflare Worker to deploy automatically on push, preview non-production branches, and show build statuses directly in a pull request.
Deepgram says DataEnclave brings AI models into customer-controlled infrastructure and helps protect sensitive data and models in use, giving enterprises another way to run real-time Voice AI.
Agencies, brands, and studios can use DIFFUSE to search for, connect with, and hire AI-native talent.
People can build free projects in Handshake’s AI Skills Studio with tools employers use. Handshake says it was built with clay, figma, GammaApp, Google, Lovable, NotionHQ, OpenAI, Replit, salesforce, and vercel, and presents it as a place to show work and get hired.
Over five days earlier in September 2026, a Chinese-speaking hacker stole hundreds of thousands of credit card details in attacks that cost $8,000 in total, Forbes reports.
Polymarket reports the forecast after Meta's new Muse AI app rose to No. 1 on Apple's U.S. App Store.
Entering one of the listed model IDs in Codex starts a query. The IDs cover named GLM, Kimi, DeepSeek, and GPT variants, with request details in B.AI’s LLM service API documentation.
Developers can use the GLM and Kimi series in the Codex client with a B.AI API key and Base URL. The API already supported GPT and DeepSeek, so it now covers four model series.
After configuring a B.AI API Key and Base URL, developers can call the GLM and Kimi model series directly in the Codex client. The API previously supported GPT and DeepSeek and now covers these four model series.
