Anthropic Launches Claude Haiku 5.5, Cutting Prices Up to 90%

Anthropic released Claude Haiku 5.5 on Wednesday, pricing its new small model at $0.10 per million input tokens and $0.50 per million output tokens for prompts up to 100,000 tokens. Haiku 4.5 cost $1 and $5. For longer prompts, Haiku 5.5 costs $0.50 and $2.50, a 50% cut.

Anthropic says about 90% of requests to Haiku 4.5 were under the 100,000-token threshold, and puts the average saving at around 75% once its new tokenizer, which uses slightly more tokens per task, is factored in.

The short-prompt price is in line with OpenAI's GPT-6 Luna, the model OpenAI began rolling out to free ChatGPT users the same day. For prompts above 100,000 tokens, Luna is cheaper.

Per-token price is not the whole cost. In Artificial Analysis testing reported by Neowin, Haiku 5.5 used about 3.5 times as many output tokens as Luna per task. It generated tokens more than twice as fast, but still took longer to finish an average benchmark task.

Anthropic is pitching Haiku 5.5 at high-volume work such as summarization, classification, database queries and subagent tasks for larger Claude models. It is the first Haiku model with an adjustable effort setting, letting developers trade cost against capability. Anthropic says Sonnet 5.5 and Opus 5.5 remain better choices for complex agentic coding.

On its own benchmarks, Anthropic reports a 72.4% score on the OSWorld 2.1 computer-use test, up from 15.7% for Haiku 4.5, and higher scores than GPT-6 Luna on most tests it published. These results are self-reported.

The model has a one-million-token context window, according to Anthropic's documentation, and is available through Anthropic's platform, AWS, Google Cloud and Microsoft Azure under the model ID claude-haiku-5-5.

Anthropic paired the launch with other price changes. Sonnet 5.5 cache reads now cost $0.10 per million tokens instead of $0.20, which the company says cuts Sonnet 5.5 costs by about 20% on most agentic work. This week, Max 5x subscribers will start getting $100 in monthly API credits, Max 20x subscribers $200, and Team plans up to $500 pooled across users. The Python and TypeScript SDKs also gained beta support for computer use and browser use.

On safety, Haiku 5.5's cybersecurity safeguards are stricter than Haiku 4.5's but looser than those on Anthropic's other recent models, according to the company. They still block penetration testing.

Anthropic did not publish independent benchmark results, say how much more the new tokenizer increases token counts, or announce changes to Haiku 4.5's availability.

Haiku 5.5 is Anthropic's first new Haiku model in nearly a year and arrives as the company prepares for an initial public offering expected before Thanksgiving.