Claude Haiku 5.5

Claude Haiku 5.5, released October 7, 2026, completed Anthropic's 5.5 generation nine days after Claude Sonnet 5.5 and fifteen after Claude Opus 5.5, and was the first new Haiku in almost a year. The headline was price. For prompts up to 100,000 tokens it cost $0.10 per million input tokens and $0.50 per million output, a tenth of Haiku 4.5's $1 and $5, with cache reads at $0.01; past 100,000 tokens the rates rose to $0.50 and $2.50, still half of Haiku 4.5's. The entry tier matched the price of OpenAI's GPT-6 Luna, and Anthropic said the model cost around 75% less to run than Haiku 4.5 on average. It was pitched at high-volume, latency-sensitive work such as classification, extraction, routing and subagent tasks, and shipped with a 1-million-token context window, 128,000 tokens of maximum output, adaptive thinking at a default of medium effort and a June 2026 knowledge cutoff.

The launch table put it far ahead of its predecessor on agentic work: 39.2% on Terminal-Bench 4.0 against Haiku 4.5's 0.0%, 72.4% on OSWorld 2.1 in its offline setting against 15.7%, and a GDPval-AA v2.1 Elo of 1620 against 735. It scored 45.9% on Humanity's Last Exam without tools and 57.4% with them, 46.4% on FrontierCode 1.1, and 46.4% on Chartography without tools against Haiku 4.5's 6.4%. Against GPT-6 Luna it led on every benchmark the table reported for both, including Terminal-Bench 4.0, where Luna scored 16.4%, and OSWorld 2.1, where it scored 48.9%; Sonnet 5.5 stayed well ahead, most of all on Terminal-Bench 4.0 at 70.6%. Anthropic reported far fewer instances of misaligned behaviour than in Haiku 4.5, with cyber safeguards stricter than Haiku 4.5's. At launch it was available in the Claude apps, through Anthropic's API as claude-haiku-5-5, and on Amazon Bedrock, Google Cloud, Microsoft Foundry and Claude Platform on AWS.