Fast Mode for Claude Opus 5.5 reaches GA as Claude Code adds client-side auto model selection

Anthropic's release notes for October 18 move Fast Mode for Claude Opus 5.5 out of beta. Developers no longer need a beta header to call it, and it is available across standard API tiers as well as through Microsoft's Azure Foundry. Fast Mode trades some compute headroom for lower latency on Anthropic's top model. GA status signals it is stable enough to build production paths on.

The Claude Code changes are aimed at day-to-day developer ergonomics. Client-side auto mode lets the CLI pick a model per request locally, routing lighter work to cheaper or faster models and reserving Opus for harder steps, without the user switching manually. Idle compaction controls let users decide how and when the agent compresses long conversation context during pauses. Startup is faster in setups with many plugins, a frequent complaint as plugin ecosystems have grown.

These updates fit a week where cost control around agentic coding is the dominant enterprise conversation. Haiku 5.5 became the default Haiku the day before at $0.10/$0.50 per Mtok, and Accenture capped central Claude funding at $29 per employee. Automatic routing between Opus, Sonnet and Haiku is Anthropic's answer to that pressure: keep developers on Claude but spend less per task. OpenAI is pushing in a different direction with parallel multi-agent execution in GPT-6 Pro, which speeds up work but can raise token usage.

Caveats: automatic model selection can make behavior harder to predict, and teams that need reproducible outputs may want to pin models. Anthropic has not published detailed latency or price deltas for Fast Mode in these notes. Not every complaint is about models, either. Ethan Mollick noted this week that users still cannot change their Claude account email without deleting the account.