Claude Haiku 5.5

Anthropic has launched Claude Haiku 5.5, a small model aimed at the kind of work that burns through tokens without needing much drama: summaries, compaction, database queries, classification, live support, and browser tasks. The pitch is simple. Keep the cheap stuff cheap, and make it fast enough that people actually use it.

The company says Haiku 5.5 is its fastest model to date, and that it works well as a subagent alongside Opus 5.5 and Sonnet 5.5 for coding work. It also comes with an adjustable effort setting, which is new for the Haiku line. In Anthropic’s framing, that gives users a choice between pushing harder for intelligence or keeping the cost down.

The benchmark table backs up the “small but not flimsy” story. On OSWorld 2.1’s offline subset, Haiku 5.5 scores 72.4%, far above Haiku 4.5’s 15.7%. On Humanity’s Last Exam, it reaches 45.9% with tools. On Terminal-Bench 4.0, it lands at 39.2%, while Sonnet 5.5 sits at 70.6%. That gap matters: Anthropic is not pretending Haiku replaces the bigger models for complex agentic coding. It is selling it as the model for narrower jobs that would have been too expensive before.

Pricing is the other half of the story. Anthropic says Haiku 5.5 is around 75% cheaper to run than Haiku 4.5 on average. For prompts up to 100,000 tokens, it lists cache reads at $0.01 per million tokens, cache writes at $0.125, input tokens at $0.10, and output tokens at $0.50. It also says roughly 90% of requests to the previous Haiku model fell into that under-100,000-token range.

And Anthropic is not stopping there. It is cutting Sonnet 5.5 cache reads in half, which it says makes Sonnet 5.5 around 20% cheaper on most agentic work. It is also giving Max and Team subscribers a monthly API credit for building on the Claude Platform. Haiku 5.5 is available now across Amazon Web Services, Google Cloud, Microsoft Azure, and the Claude Platform.