OpenAI's GPT-6 Astra and Sol now run ~50% faster, partner apps too
The ~50% boost is on by default and reaches apps using Sign in with ChatGPT - plus a Codex Windows MCP fix and CodeMender locking out CI/CD edits.
Copy markdown
Astra and Sol stream ~50% faster, no price change
OpenAI core-products head Tibo Sottiaux says GPT-6 Astra and GPT-6.1 Sol now stream about 50% faster by default - reports peg it at roughly 30 to 50 tokens/sec - with no change to pricing or plan limits. Lower latency lands across ChatGPT, Codex, and the surfaces you already use.
Your agent apps inherit the speedup for free
The gain extends to third-party apps using 'Sign in with ChatGPT' - OpenCode, Pi, Amp, and Devin were named - so your agent tooling gets faster inference with no SDK change, billed against your existing ChatGPT plan under per-app weekly caps.
The catch: it's OpenAI's own number
No independent benchmark shipped with the claim - the 50% is OpenAI's own, framed as how the models 'feel.' Time your own p50/p95 before you re-budget latency-sensitive flows. It's billed as Day 1 of OpenAI's 28-day ship sprint, so expect more daily changes.
Codex CLI 0.160.1 unbreaks Windows MCP launches
If you run remote stdio MCP servers from Codex on Windows, 0.160.1 now preserves SYSTEMROOT, TEMP, and TMP when launching them - a backport to the 0.160 line that fixes MCP startup in mixed Unix/Windows setups. Grab it with npm i -g @openai/codex@0.160.1.
CodeMender v0.12.0 stops touching your CI/CD
Google's security agent now refuses to modify CI/CD pipelines, build definitions, and static-analysis configs outside a vulnerability's scope - a real guardrail if you let it auto-patch. It also added hidden-file scanning (--include-hidden) and raised the scan size cap from 500KB to 2MiB.