Claude Haiku 5.5 Review

Haiku 5.5 is the third Claude 5.5 model, released on October 7, 2026, and Anthropic sells it as the cheapest, fastest and most capable small model it has made. The price is the headline. For prompts up to 100,000 tokens, which Anthropic says covers about 90 percent of Haiku 4.5 requests, it costs 10 cents in and 50 cents out per million tokens, a tenth of Haiku 4.5's $1 and $5, with cache reads at a cent. Above 100,000 tokens it rises to 50 cents and $2.50. A new tokenizer, shared with Sonnet 5.5 and Opus 5.5, uses slightly more tokens per task, and Anthropic's figure of about 75 percent saved on average already counts it.

The capability jump is as large as the price cut. On Anthropic's table it scores 39.2 percent on Terminal-Bench 4.0 where Haiku 4.5 scored zero, 72.4 percent on an offline OSWorld 2.1 subset against 15.7, and 1620 on GDPval-AA against 735, and it beats OpenAI's GPT-6 Luna, its direct rival at the same price, on every row the two share. It is also the first Haiku with adjustable effort, so a team can trade cost against intelligence call by call. Early customers describe the practical version: Asana cut task completion latency by more than 30 percent, and Box measured an 11 point gain over Haiku 4.5 with latency roughly halved.

Its limits are stated plainly. Sonnet 5.5 beats it on every benchmark in the same table, and Anthropic says the bigger models remain the better choice for complex agentic coding. Haiku 5.5 is for narrowly scoped work, such as summaries, classification, compaction, database lookups, live support and browser use, or for running as a subagent under Opus or Sonnet. Its cyber safeguards are stricter than Haiku 4.5's and still block penetration testing. For high volume work it is the best value anyone sells, which earns repetitive tasks an S, and it is an A.