Issue 3: Fable 5.1 lands, Claude Code limits shrink
Fable 5.1 and GPT-6 Astra arrived alongside changes to how developers get access: Claude Code’s temporary allowance boost shrinks on 14-09-2026, and OpenAI proposes ending Cursor’s direct model access on 12-11-2026. Meta’s Muse Spark 1.3 matches leading models on benchmarks, with a large price discount for customers who opt into training. Practitioners described agents handling open source maintenance and exploit probes arriving minutes after a patch was shared.
- Muse Spark 1.3 matches GPT-5.6-SolMeta matches leading models on benchmarks and offers a discount above 90% for customers who opt into training; the price comes with a data choice
- OpenAI plans to end Cursor access in NovemberOpenAI proposes ending Cursor's direct model access on 12-11-2026 after its SpaceX acquisition; developers have notice, but future models are excluded
- Fal H3 Max Live breaks the videogen barrierFal demonstrates video generation faster than playback, with viewers directing an ongoing stream through chat; a working example of live generated video
- PRs not welcome: agents as software factoriesVercel's AI SDK and Astro use agents to triage and fix issues; Flue and tldraw close external PRs and ask contributors to start with issues instead
- Understanding ChatGPT WorkA close read of OpenAI's cloud agent product: sandboxed execution with internet access, a shared filesystem, headless Chrome and sub-agent sessions
- A rumour of a bug is enough to find an exploitAn OCaml maintainer reports exploit probes within minutes of a patch being shared for discussion; days used to be the norm
- Claude Fable 5.1 and Mythos 5.1 launchFable 5.1 is generally available; Mythos stays restricted to trusted programs. Cache reads cost 75% less than Fable 5, making repeated context cheaper
- Claude Code's temporary limit boost shrinksThe linked discussion quotes a move from 150% to 125% of baseline on 14-09-2026, about 17% below the temporary allowance; plan long runs around it
- Codex CLI makes GPT-6 Astra its default modelVersion 0.153.4 bundles Astra as the default when no model is configured, so an unpinned setup changes model on upgrade
- GPT-6 Astra reaches ChatGPT and the APIThe rollout covers paid ChatGPT tiers and the major cloud APIs, with faster computer use and new Codex context memory