We use cookies to measure how the site is used and how our ads perform (Google Analytics and Ads, Microsoft Clarity, and the Meta, OpenAI and X ad pixels). “Essentials only” turns off advertising cookies everywhere, and analytics too in the EEA, UK and Switzerland. Privacy Policy
Tool Retrieval is in preview in the Gemini Interactions API: add a tool_search tool, set defer_loading on the long tail, and the server loads parameter schemas on demand. What that changes for tokens, prefix caching, MCP catalogs and debugging — and where the saving does not show up.
Job listings spotted by Bloomberg show Apple building an autonomous agent that browses, codes, controls apps and coordinates sub-agents — the last major platform holder to enter the computer-use race.
OpenUI 1.0, the MIT-licensed generative UI runtime, adds custom functions, custom actions and library.extend to its streaming UI language, plus an Agent Skill so Claude Code, Codex and Cursor can scaffold generative-UI apps. Vue, Svelte and Angular stay on 0.4.
Cloudflare and OpenAI's research pilot used network signals from ~1% of Cloudflare's network to help ChatGPT Search discover hundreds of millions of previously unknown pages, plus spam filters for nonexistent pages, redirect abuse, and single-compromised-page flags.
Docker's Harbor eval harness now runs every task inside Docker Sandboxes MicroVMs with staged network policies, while Docker's agent security playbook supplies the threat model and LangChain's new guide covers what to actually measure.
At its first Config in India, Figma launched Specs, a Markdown editor for design rules that agents can consult as context, moved code layers from private to public beta, and added more than 20 tools to its MCP support, TechCrunch reports.
On its Q3 2026 earnings call, Bank of America CEO Brian Moynihan disclosed 1,000 contractors and hundreds of millions in costs tied to patching flaws Anthropic's Mythos model exposed — one of the first concrete megabank tallies of frontier-model dual-use risk.
Ling-3.1-flash pairs 560B total and 25B active parameters with a 1M-token native context window, an MIT license badge on its model card, day-0 SGLang and vLLM support, and a Vercel AI Gateway model ID callable from Claude Code and Codex. Verified specs, recipes and chronology.
GPT-5.5 retired from Codex for ChatGPT-authenticated sessions on October 14, 2026 at noon PDT. Here is what changes now, which plan moves to gpt-6.1-sol or gpt-6-luna, and why API-key sessions keep working.
Federal prosecutors say an OpenAI engineer hid an encrypted audio recorder in an executive conference room, with chatbot prompts tracing the alleged plan; OpenAI says model weights, source code and customer data were not accessed.
Kylon's company harness puts its built-in Operations agent Ryan alongside user-brought Claude Code, Codex and Devin, with sandboxed cloud runtimes, draft-first approvals for email and line X sends and credit metering by room, agent and person.
Anthropic's Oct. 14 desktop refresh turns Claude Code into a control room: start parallel sessions from the sidebar, isolate each in its own Git worktree, watch subagents in the tasks pane, ask side questions off-thread, and move long runs to the cloud.
LangChain's Oct. 14 Deep Agents guide explains five middleware techniques for recoverable context engineering: offloading, summarization, skills, subagents, and file-based memory that survives model switches.
Modal announces a $1 billion annualized run rate and a $750M Series D at a $15.75B post-money valuation led by Accel, with Sandboxes alone at a $300M run rate. What agent builders should plan around now.
Google Research opened Deep Review, its multi-agent document verification system piloted on 16,000+ conference manuscripts, to everyone. Two free reviews a week, PDF and text input, no training on uploads, one-year storage, and a Gemini API path for CI pipelines.
Mistral buys New Material, a six-year product and design studio, and folds it into a new Solutions Design function that co-develops with enterprise and government buyers.
Hinton, Bengio, Russell, and Rivest join 400+ scientists asking for high-assurance containment, an international standards body, and mandatory public incident reporting after disclosed sandbox escapes.
The same day Google took Gemini 4 Argon fully live, Cursor added it to its model picker at $2/$10 per million tokens; Vercel's AI Gateway also opened it to Claude Code, Codex and Cursor users through one gateway key.
Workday made its Trusted Agent Platform headless at Workday Rising: an MCP server with 275 read and 54 write tools, SQL access to live data, and a shared Agent Harness runtime are all generally available, scoped to the signed-in user's own permissions.
Liquid AI releases the final LFM2.5 model with open weights, 131K context, and tool-calling post-training. It decodes 184 tok/s on an Apple M5 Max and leads AutomationBench by 3× over the next-best MoE.
A seven-person team built Scion, an always-online client-server version control system, and migrated Anthropic's entire monorepo from Git in five months after daily merges surged from 2,500 to over 30,000.
Google's frontier model is live today: paid API lanes get gemini-4-argon, Antigravity stays Ultra-only, the Gemini app needs Pro, and the $2/$10 intro rate per million tokens doubles on November 30, 2026.