Claude Haiku 5.5 Review: Features, Benchmarks & Pricing

Claude Haiku 5.5 is Anthropic's newest lightweight model, released on October 7, 2026, nine days after Claude Sonnet 5.5. It is Anthropic's cheapest and fastest current model, priced at just $0.10 per million input tokens for most requests — a tenth of what Claude Haiku 4.5 costs per token — while scoring many times higher on coding and computer-use tests.

This Claude Haiku 5.5 review covers what changed, what the benchmark numbers show, how its two-tier pricing works, and where its limits sit. It also covers access: Claude Haiku 5.5 is not yet integrated into Chat Smith, but Claude Haiku 4.5 and other Claude models already are.

Claude Haiku 5.5 is the small, fast member of Anthropic's Claude 5.5 family, built for high-volume, repetitive work: summarizing documents, classifying information, querying databases, and handling smaller assignments for more capable agents. Anthropic calls it its best lightweight model yet, ahead of Haiku 4.5 in coding, computer use, and knowledge work.

It supports a 1-million-token context window, up to 128,000 tokens of output, text and image input, and adjustable effort levels from low to max. It is available as claude-haiku-5-5 through the Claude API and major cloud platforms.

Haiku sits below Sonnet and Opus in Anthropic's lineup. If you need deeper reasoning on harder tasks, our Claude Sonnet 5 review covers the mid-tier model already available in Chat Smith.

Four areas show the biggest gains over Claude Haiku 4.5.

On Terminal-Bench 4.0, which tests autonomous command-line tasks, Haiku 5.5 scores 39.2%, up from 0.0% for Haiku 4.5 and well ahead of GPT-6 Luna at 16.4%. Independent tester Vals AI also ranks it third of 110 models on Vibe Code Bench v1.1 with 90.44%, an unusually strong result for a small model.

On the OSWorld 2.1 offline subset, which measures desktop tasks such as navigating apps and filling in forms, Haiku 5.5 reaches 72.4%, compared with 15.7% for Haiku 4.5 and 48.9% for GPT-6 Luna. That makes it a realistic option for lightweight agents that need to operate software, not just write text.

Haiku 5.5 is designed for jobs that run thousands of times a day: summaries, tagging, extraction, quick replies, and support drafts. In one hands-on comparison, it and Haiku 4.5 both completed every test task, but the Haiku 5.5 run cost about $0.01 against $0.25. For work like condensing reports with an AI summarizer or drafting replies with an AI email writer, that efficiency adds up fast.

Anthropic positions Haiku 5.5 as a cheap worker inside larger agent systems, handling small subtasks for Sonnet or Opus. Its effort setting lets you trade cost for depth: at low or medium effort it stays extremely cheap, while max effort used around eight times more output tokens than medium in one independent test.

The table below uses Anthropic's launch figures, comparing Haiku 5.5 with its predecessor and with GPT-6 Luna, OpenAI's model at the same price. Prices are API rates per million tokens.

The pattern is clear. Haiku 5.5 beats both on every shared benchmark, matches GPT-6 Luna's price for most requests, and costs a fraction of Haiku 4.5. The one catch is long prompts, where Haiku 5.5's per-token price rises fivefold.

Claude Haiku 5.5 is priced by prompt length. For prompts up to 100,000 tokens, it costs $0.10 per million input tokens and $0.50 per million output tokens — 90% less than Haiku 4.5. Above 100,000 tokens, the rates rise to $0.50 and $2.50. Anthropic says about 90% of previous Haiku requests fall under the threshold, and estimates that workloads cost about 75% less on average once its newer tokenizer, which counts the same text as roughly 30% more tokens, is factored in.

• Developers: available as claude-haiku-5-5 on the Claude API, Amazon Bedrock, Google Cloud, and Microsoft Foundry.
• Related change: Anthropic also cut Claude Sonnet 5.5's cache-read price to $0.10 per million tokens at the same time.
• Chat Smith: Claude Haiku 5.5 is not integrated yet. You can use Claude Haiku 4.5 and other Claude models today, with higher usage limits on Chat Smith Pro at $9.99 per month or $39.99 per year.

The low price applies only to shorter prompts. Once a request passes 100,000 tokens, Haiku 5.5 costs five times as much per token, so long-document workloads see a much smaller saving over Haiku 4.5.

It is still a small model. Claude Sonnet 5.5 leads it on every benchmark Anthropic published — 70.6% against 39.2% on Terminal-Bench 4.0, for example — and on Vals AI's broader index it ranks 16th of 45 models, with weak results on specialist legal tasks.

Max effort can also erase much of the savings because of its heavy token use. Start Haiku 5.5 at low or medium effort, and for anything that needs more depth, compare it with alternatives such as Claude Sonnet 5.

Claude Haiku 5.5 is not integrated into Chat Smith yet, but several Claude models already are. Pick one from the model list and start chatting on Android, iOS, or in the browser, with no Anthropic account needed.

Claude Haiku 4.5 is the closest option: Anthropic's previous fast model, built for quick answers, summaries, and lightweight coding help. When a task needs more careful reasoning, switch to Claude Sonnet 5, and Claude Sonnet 4.6 remains available if you have prompts tuned for it. Haiku 5.5's biggest gains show up mainly in agentic coding, computer use, and very high-volume API workloads.

Because Chat Smith also includes models from OpenAI, Google, xAI, and DeepSeek, you can put the same prompt to a Claude model and its rivals and keep the answer that holds up. Chat Smith Pro adds higher usage limits and more premium models for $9.99 per month or $39.99 per year.

Claude Haiku 5.5 resets expectations for small models: multi-fold gains over Haiku 4.5 on coding and computer use, stronger results than GPT-6 Luna at the same price, a 1M-token context window, and a 90% lower per-token price for most requests. Its main caveats are the steeper rate for long prompts and the gap that remains to Sonnet-class models.

Until it reaches Chat Smith, the Claude models already in the app are an easy way in. You can try them now in Chat Smith, an AI chatbot that puts Claude next to the latest models from OpenAI, Google, xAI, and DeepSeek.