Fast mode retires on Claude Opus 4.8 Nov 30 — migrate to Opus 5.5
Anthropic put a 30-day clock on fast-mode calls to the 4.8 API; after Nov 30 they error out. Plus: Claude wrote the Rust rewrite now behind the Messages API.
Copy markdown
Fast mode on Opus 4.8 is deprecated
Anthropic's API release notes flagged `speed: "fast"` on `claude-opus-4-8` for removal on Nov 30, 2026 — after that, fast-mode requests to the 4.8 endpoint return an error. Standard-speed calls keep working, but any pipeline that hardcodes fast mode on 4.8 breaks that day. The clean fix is Opus 5.5, which keeps fast mode (and runs it on Anthropic-managed Google Cloud) and costs ~40% less to run than Opus 5 — so pin your model version and migrate now rather than eat a surprise error in a month.
Claude wrote the Rust rewrite now behind the Messages API
On the Latent Space podcast — resurfacing via a viral clip this week — Anthropic engineers said Claude wrote the Rust port that now serves the production Messages API, with humans designing the system and shadow-testing old vs. new on real traffic. The payoff they cite: roughly 40–50% less CPU and 10–20% faster time-to-first-token, already rolled out. Take the numbers as podcast figures, not a published benchmark, but the lesson lands: the "never do the big rewrite" rule looks a lot weaker when an agent does the porting.