Anthropic released Claude Haiku 5.5 today, calling it the cheapest, fastest, and most capable small model it has shipped. It costs about 75 percent less to run than Claude Haiku 4.5 on average, and it scores 72.4 percent on the OSWorld 2.1 computer use benchmark, up from 15.7 percent.
Anthropic describes Claude Haiku 5.5 as a model for high-volume, cost-sensitive work such as summaries, compactions, database queries, and classification. It is the third model in the Claude 5.5 family, after Claude Opus 5.5 and Claude Sonnet 5.5, and it is meant to run as a subagent next to both on coding jobs.
Claude Haiku 5.5 Costs About 75 Percent Less Than Claude Haiku 4.5
Anthropic prices Claude Haiku 5.5 by prompt length. Prompts up to 100,000 tokens, which made up around 90 percent of requests to Claude Haiku 4.5, cost 90 percent less. Longer prompts cost 50 percent less.
The 75 percent average also accounts for a new tokenizer, which uses slightly more tokens per task, similar to Claude Sonnet 5.5 and Claude Opus 5.5. Here is the price per million tokens for prompts up to 100,000 tokens:
- Input tokens: $0.10 for Claude Haiku 5.5, against $1.00 for Claude Haiku 4.5 and $2.00 for Claude Sonnet 5.5
- Output tokens: $0.50, against $5.00 and $10.00
- Cache reads: $0.01, against $0.10 and $0.10
- Cache writes: $0.125, against $1.25 and $2.50
Past 100,000 tokens, input costs $0.50 per million, output costs $2.50, cache reads cost $0.05, and cache writes cost $0.625.
Anthropic is also cutting the price of cache reads on Claude Sonnet 5.5 in half, from $0.20 to $0.10 per million tokens. Cache reads make up a large share of token use in agentic work, so the company says Claude Sonnet 5.5 now runs around 20 percent cheaper on most of those tasks.
Introducing Claude Haiku 5.5: the cheapest, fastest, and most capable small model we’ve ever released.
On average, it costs around 75% less to run than Claude Haiku 4.5. https://t.co/jm06cZJfkV
— Claude (@claudeai) October 7, 2026
Claude Haiku 5.5 Jumps From 15.7 to 72.4 Percent on OSWorld 2.1
Anthropic compared Claude Haiku 5.5 with Claude Haiku 4.5 and OpenAI’s GPT-6 Luna. Here is how they compare in its table:
- OSWorld 2.1 computer use, offline subset: 72.4 percent, against 15.7 percent for Claude Haiku 4.5 and 48.9 percent for GPT-6 Luna
- GDPval-AA v2.1 knowledge work: 1620, against 735 and 1437
- AA-Briefcase v1.1 knowledge work: 1578, against 614 and 1336
- Terminal-Bench 4.0 agentic coding: 39.2 percent, against 0.0 percent and 16.4 percent
- FrontierCode 1.1 agentic coding: 46.4 percent, against 42.4 percent for GPT-6 Luna
- Humanity’s Last Exam: 45.9 percent without tools and 57.4 percent with tools, against 10.2 percent and 18.7 percent for Claude Haiku 4.5
- Chartography chart reasoning with no tools: 46.4 percent, against 6.4 percent and 29.1 percent
Claude Haiku 5.5 is the first Haiku model with an adjustable effort setting, so users can choose between lower cost and higher intelligence the same way they can with the larger models.
Anthropic Positions Claude Haiku 5.5 for Subagents, Summaries, and Browser Use
Claude Sonnet 5.5 and Claude Opus 5.5 remain the better picks for complex agentic coding, with Claude Sonnet 5.5 scoring 70.6 percent on Terminal-Bench 4.0. Claude Haiku 5.5 is aimed at narrower jobs that were too expensive to run before, such as compaction, summarization, and subagent work. Rogo uses it that way, sending a Claude Haiku 5.5 subagent into a 10-K to pull the segment revenue line while a larger model builds the deck.
It is also Anthropic’s fastest model at each model’s standard speed, though it runs less quickly than the Opus models in Fast Mode. Anthropic says that speed suits it to live customer support and browser use.
The cybersecurity safeguards on Claude Haiku 5.5 are stricter than those on Claude Haiku 4.5 and somewhat looser than those on other recent models. They allow a wider range of defensive tasks than the safeguards on Claude Sonnet 5.5, but they still block penetration testing and other techniques more likely to be used by attackers. Biology safeguards match Claude Sonnet 5, Claude Sonnet 5.5, and Claude Opus 5. Organizations that need broader access can apply to the Cyber Verification Program or the Life Sciences Verification Program.
The system card reports far fewer instances of misaligned behavior than Claude Haiku 4.5, along with a lower willingness to cooperate with misuse.
Claude Haiku 5.5 is available now on all platforms, including Amazon Web Services, Google Cloud, and Microsoft Azure. Developers can call it with the model ID claude-haiku-5-5, and Anthropic has a migration guide for moving over from Claude Haiku 4.5.
Anthropic is also adding beta support for computer use and browser use to its Python and TypeScript SDKs. On top of that, Claude Max and Team subscribers get a new monthly API credit for building on the Claude Platform, rolling out this week. Claude Max 5x subscribers get $100 per month, Claude Max 20x subscribers get $200, and Team subscribers get up to $500 pooled across their users, usable on any Claude model. Details are in the Help Center.
