AI models, services and robotics
Through October 15, work that follows creating or editing a doc, slide deck or design artifact in Claude counts at half rate against the five-hour session limit. The offer covers Pro, Max and Team plans, while the weekly limit, Claude Code and the API are unchanged.
The blog of OpenAI's Strategic Futures team has published the first essay in its Next Economy series, arguing that the value of superintelligence may come less from discovery than from execution and coordination. Citing research showing that research productivity has fallen 41-fold since the 1930s, it sketches two possible futures: a civilization of depth and a civilization of width.
MongoDB released Atlas Agent Engine in public preview at its September 29 Investor Day, bundling AI agent runtime, memory, and governance into one platform. The company's pitch: it isn't tied to any single model or cloud, and runs against customers' existing Atlas commitments.
Anthropic has released mods, TypeScript functions that change how Claude Code behaves and what it shows. Mods install as plugins and run unsandboxed, with the same access to your machine as Claude Code itself.
Meta has launched Ray-Ban Meta Gen 3 in India and named Ranveer Singh as its first brand ambassador there. It also previewed the 43-gram Ray-Ban Meta Audio and plans to bring Muse to its AI glasses.
British bank Barclays is extending Anthropic's Claude across the company and aims to have 50% of its developers using Claude Code by the end of 2026. An internal search tool used by 16,000 employees and the sorting of 120,000 emails a day already run on Claude.
A flu hospitalization forecasting model Google built with its AI research tool ERA posted the best score among 39 models in the CDC's season evaluation. Its margin over second place was 0.02, and state-level rankings were mixed.
The feature gathers the Sites a user built in ChatGPT and their most-used plugins on one page. Personal profiles are private by default, while business profiles are shared by default.
Google released its new frontier model to cyber defenders first. It raised the output token limit to 1 million and set no date for general availability.
Anthropic will run Claude Founder House, an in-person event for founders, in San Francisco on October 6–8 and in Stockholm on October 14. The program includes talks, workshops, office hours with the Applied AI team and live demo voting.
OpenAI said it disrupted a coordinated distillation campaign aimed at extracting protected reasoning from its models, and attributed a core cluster to individuals associated with Moonshot AI, the developer of Kimi. Over two days in July, more than 4,000 users sent 16,000 requests.
Anthropic has opened claude.dev, a developer hub that gathers engineering deep dives and Claude Code and API guides. Its flagship posts are the record of a sprint that made claude.ai about 3x faster in two weeks and a method for automating eval design.
OpenAI announced a partnership with America's SBDC, the association of U.S. small business development centers, to train about 150 advisors who will run AI workshops for at least 1,000 small businesses. A report released alongside it says two-thirds of small-business output tokens now come from agentic products.
Google DeepMind has unveiled SynthID Bio, which embeds watermarks in the sequences and 3D structures of AI-designed proteins. In wet-lab tests on three target proteins, the watermarked designs kept their function, and the company is releasing the code and weights.
At DevDay 2026, OpenAI launched Ultrafast, which runs GPT-6 Astra at up to 300 tokens per second. API pricing is 6x the standard rate, and in ChatGPT Work and Codex it is available only on the new Pro 500 plan and Enterprise.
OpenAI has released the Decisions API in limited preview, letting GPT-6 Luna pick one answer from a list defined by the developer. In the official demo, it took 150 milliseconds per request, about 10 times faster than the 1.6 seconds of the existing Responses API.
Google Research unveiled Diffusion Controller, a control framework that keeps the original image generation model frozen and uses a small side network to correct the direction of generation. A gray-box version with 12 million parameters achieved higher preference win rates than LoRA.
Anthropic found that Zhipu AI's open-weight model GLM-5.3 can build cyber exploits at a level close to Claude Mythos Preview. Simple techniques bypassed its safeguards between 64% and 100% of the time.
Running November 9-10 at Seoul's COEX, the event is expected to draw about 2,000 developers, startups, enterprise teams, and researchers
LLM token-usage dashboard for easier management... Operations and observability upgraded with data-drift detection and more
"No more tools that connect but don't run! Breaking down the wall of complex MCP integration"
OpenAI released GPT-6.1 Sol, which it says comes close to GPT-6 Astra at a fifth of Astra's standard token prices. Its system card addendum classifies the model as Critical in cybersecurity.
OpenAI introduced dots, always-on agents powered by GPT-6 Astra, at DevDay. Each dot has its own cloud computer, connects to more than 4,000 apps and works around the clock.
Anthropic released a developer guide for teams moving to Claude Sonnet 5.5. It lays out five breaking changes and says the old setting for turning thinking off and forced tool calls now return 400 errors.
