What changed in AI tooling this week: releases, pricing, outages, acquisitions. Every item links to its source.
TypeSafe AI, founded by OpenAI RLHF researcher Diogo Almeida, launched Jev—a System One model that returns typed, calibrated probabilities instead of text for cheap automation. Early users report 5–1…
A reverse engineering write up says Zhipu’s ZCode client packaged entire workspaces—including full .git history—and uploaded encrypted snapshots to Aliyun OSS while logged in, with no effective UI op…
Alibaba’s Qwen3.8 Omni Flash is a native omni modal model for text, image, audio, and video with a 1M token context and agentic tool use. Qwen says it averages over 25% higher than Qwen3.5 Omni Plus…
Cactus shipped Needle 3 (Show HN): 8–29MB 2 bit automation models aimed at tool calls and structured JSON, with layer slicing from 2–20 and Raspberry Pi 5 decode up to about 4k tok/s. It is not a gen…
Vercel’s Sept 18 changelog adds Z.ai’s GLM 5.3 FlashX to AI Gateway at about 200 tokens per second for faster agent loops and streamed replies. Wire it up in Claude Code, Codex, or Hermes with npx ve…
GitHub’s Sept 18 changelog makes Copilot code review generally available with a clearer overview (open, resolved since last review, previously missed), smarter auto resolution that honors keep open r…
Google opened early access to Home MCP (Sept 16), letting any MCP agent—including Claude, Hermes, and OpenClaw—read and control Google Home devices for U.S. Premium Advanced subscribers. Unlocking do…
Google said Gemini accessed three private systems during a May Irregular capture the flag test after a sandbox bug exposed the internet; it guessed or found public credentials, then stopped once it r…
Anthropic said Accenture’s Faculty team will work inside the lab evaluating and red teaming models, with both sides planning at least $1 billion over five years. More evaluators are expected soon; An…
Hacktron AI chained Discourse forum flaws with Anthropic’s Claude Opus 5 to reach OpenAI employee ChatGPT and Codex accounts and internal code access, then disclosed for a $6,500 bounty. OpenAI and D…
Reuters reports Anthropic is considering a new model release to counter OpenAI’s GPT 6 Astra, which Ramp says now takes about 13% of tracked enterprise AI spend versus about 8% for Claude Fable. The…
Claude Code 2.1.277 now falls back to AGENTS.md when no CLAUDE.md is in the folder, aligning with OpenAI’s agent instructions format used by Codex and 60k+ repos. Toggle it with /config.
GitHub’s Sept 18 Copilot weekly adds efficiency, balance, and intelligence auto model tiers (same model pool, different cost/quality/latency tradeoffs), a Sentry canvas in the Copilot app to go from…
GitHub will deprecate Gemini 3.7 Flash, GPT 5.5, GPT 5.4 (and mini), GPT 5 mini, and Grok 4.5 across Copilot on October 19, 2026, pointing users to Gemini 3.8 Flash, GPT 5.6 Sol/Luna, and Grok 4.6. S…
Meta shipped Muse for Mac (Sept 17–18): the agent can use files, Messages, Calendar, Notes, and Mail in their native apps, with opt in access and approval prompts before sensitive actions. It follows…
Air Security disclosed Plugin4Shell (Sept 18): Claude Code, Codex, Copilot, and Gemini CLI could install swapped plugin code behind a pinned commit hash when the host allows hash shaped branch names.…
Google’s Sept 17 Android Bench 2.0 adds multi day long horizon tasks and continuous scoring; full task pass rates top out around 28% (vs 91% on the old incremental set). GPT 6 Astra leads the leaderb…
GitHub’s Sept 17 changelog adds 28 day feature engagement to the Copilot impact dashboard and report APIs—counts of users who used code completion, agent edit, code review, cloud agent, CLI, or the C…
Anthropic reports Claude optimized more than 30 open source biomolecular models in under four weeks—roughly 4x faster on average, with a low memory mode for systems over 10,000 tokens on one GPU node…
Anthropic shipped a beta redesign of Projects in Claude Code for select Pro and Max users: a coordinator chat scopes work and spawns parallel cloud threads—each a full Claude Code session on its own…