Anthropic releases Claude Haiku 5.5 with effort settings and adaptive thinking

Anthropic released Claude Haiku 5.5, its first Haiku-tier update in roughly a year, positioning the model as a subagent paired with Opus 5.5 or Sonnet 5.5 for high-volume, cost-sensitive tasks such as summaries, compactions, and database queries. Anthropic described it as the cheapest, fastest, and most capable small model it has released, costing about 75% less to run than Haiku 4.5 on average.

The model introduces Anthropic’s effort settings and adaptive thinking, with tiered pricing that halves cache-read costs for Sonnet 5.5 and introduces new API credit tiers for subscribers. It is available on the Claude Platform and in Claude Code, with same-day SDK updates adding built-in computer-use and browser-use toolsets for Python and TypeScript.

Haiku 5.5 features a 1M-token context window, up from 200K for Haiku 4.5, and supports text and image input with text output. Anthropic’s tiered pricing structure charges $0.10/$0.50 per 1M tokens for input/output under 100K tokens and $0.50/$2.50 above that threshold, with cache reads priced at $0.01/$0.05 per 1M tokens.

Independent evaluations by Artificial Analysis report Haiku 5.5 scoring 43 on the Artificial Analysis Intelligence Index at max effort, up 26 points from Haiku 4.5, and slightly ahead of GPT-6 Luna (38) and comparable to Kimi K3 (44). On Terminal-Bench 4.0, Haiku 5.5 scored 33%, up from 0% for Haiku 4.5, and on AA-Briefcase it achieved 1578 Elo, ahead of Kimi K3 and GLM-5.3 Flash.

Anthropic’s pricing and positioning reflect a strategy to compete directly with OpenAI’s GPT-6 Luna, with the company cutting prices on Sonnet 5.5 and introducing new API credit tiers the same day as the Haiku 5.5 release. Partners including Cursor, GitHub Copilot in VS Code, Devin, Arena, and OpenDocRouter announced same-day support for the model.