NEWS

How OpenAI built GPT-Live: a full-duplex voice model with no turn detector, Go on the media path, stateful handoffs, and the new WARP WebRTC transport.

Qwen 3.8 Max is a 2.4T-parameter MoE with 95B active, a 1M context window and $2/$6 per million tokens on QwenCloud, with open weights promised next week.

The EU's final DMA decision gives third-party AI assistants the Android access Gemini has, from DSP wake words to screen automation. What it unlocks and when.

Karpathy gave Opus 5 one Lord of the Rings paragraph and a 1M-token budget for a three.js render. What the run shows about agentic work and self-verification.

Cursor removed dollar costs from its Usage page on July 31: self-serve plans see tokens only and the dashboard API zeroes cost fields, history included.

Gemini 2.5 Pro and Gemini 3 Flash are deprecated in GitHub Copilot. What GitHub suggests instead, the October follow-up deprecations, and how to migrate.

OpenAI's Astra model produced results on ten decade-old math problems, including Erdős problems, each with a Lean 4 certificate. What shipped and why.

A ripgrep musl segfault in huge multi-threaded searches traced to a suspected Linux 7.0 kernel race. How it was pinned, plus the one-line fix candidate.

Vercel AI Gateway spend budgets now cover teams and projects, with hard dollar limits, 50/75/100% email alerts and CLI defaults. How the three scopes compose.

Vercel MCP now serves the stateless 2026-07-28 spec and the 2025 protocol from one endpoint via mcp-handler 2.x. What it means for server authors and clients.

AGENTS.md and CLAUDE.md configuration smells: a SCAM 2026 study of 100 top repos catalogs six, led by Lint Leakage and Context Bloat. What to fix in yours.

AgentS4D tests runtime safety for Claude Code, Codex, OpenClaw and Hermes across 6,560 runs: most unsafe runs still completed. Completion is not safety.

AI session portability: sealed reasoning, opaque compaction and hidden subagent messages are replacing portable transcripts. Five tests and seven rules for it.

Blind resampling beats self-repair in small code models: a placebo-controlled MBPP+ study finds retrying from scratch beats feeding failed code back.

CAPA benchmark: coding agents that read a user's past sessions resolve ambiguous requests with far fewer clarifying questions, tested across 12 models.

cdnjs now runs entirely on Cloudflare Workers, Workflows, R2, and Queues. How the publishing pipeline was rebuilt and the two platform limits it raised.

Change2Task turns merged pull requests into verified, executable coding agent tasks that stay current with the codebase, across five task families.

Coding agents rarely read open source AI contribution rules on their own and never refuse AI-banned repos. RepoComplianceBench shows what prompts can fix.

AGENTS.md and context files do not measurably change coding agent correctness in a 288-run Claude Code and Codex ablation. Failures were implementation skill.

Gemini Robotics ER 2 is public in the Gemini API. It watches live video to track progress, orchestrates VLA models as tools, and coordinates multiple robots.