Claude Code v2.1.316 makes Haiku 5.5 the default model for Pro subscribers

The update changes what most Pro-tier Claude Code users run by default. Haiku 5.5 is Anthropic's fastest and cheapest model in the current family, and it now handles generation, refactoring and debugging unless the user manually switches up.

The practical effects:

- Lower latency in interactive loops.

- More headroom under usage limits, since Haiku consumes less of the plan's budget per task.

- The /model command still allows switching to Opus- or Sonnet-class models for harder work.

Read with the rest of the week, it is clearly a cost move as much as a UX one. OpenAI is merging ChatGPT and Codex into a single token pool. xAI says Grok agent serving is unsustainable. Anthropic has filed for an IPO, where serving margins will be scrutinized. Defaulting lower-tier subscribers to a smaller model is the gentlest lever available, gentler than cutting quotas. It also pre-empts the backlash seen elsewhere.

Competitively, Claude Code now faces DeepSeek's MIT-licensed Super-Agent, explicitly built to rival it, and Microsoft's unified Copilot with Code built in. Developer goodwill is already strained by the new rule restricting recreation of specific video games and by reports of account blocks.

The caveat is quality. Developers in Hacker News threads praise Claude Code for large legacy migrations, such as 50K lines of ColdFusion moved to Java in two weeks. Those are the workloads where a smaller default model may underperform. Watch for complaints that Haiku defaults degrade long-horizon refactors, and for whether Max-tier defaults change next.