Claude Haiku 5.5 Cuts Small-Model Prices by Up to 90% and Adds an Effort Setting

Anthropic released Claude Haiku 5.5 on October 7, 2026, describing it as “the cheapest, fastest, and most capable small model we’ve ever released.” The company says it costs around 75% less to run than Haiku 4.5 on average, and that it is the first Haiku-class model to ship with an adjustable effort setting. Developers can call it on the Claude Platform as claude-haiku-5-5, and Anthropic says it is available on all platforms, including Amazon Web Services, Google Cloud and Microsoft Azure.

Anthropic positions Haiku 5.5 for high-volume, cost-sensitive work: summaries, context compactions, database queries and classification. It also pitches the model as a subagent that pairs with Opus 5.5 and Sonnet 5.5 on coding tasks. Because the company says it is its fastest model to date at standard speed, it is recommending Haiku 5.5 for latency-sensitive jobs such as live customer support and browser use. A footnote adds that it still runs slower than Opus models in Fast Mode.

How the 75% figure is built

The headline saving comes from a two-tier price list. For prompts up to 100,000 tokens, Haiku 5.5 costs $0.10 per million input tokens and $0.50 per million output tokens. Above that threshold, the rates rise to $0.50 and $2.50. Haiku 4.5 cost $1.00 and $5.00. Cache reads run $0.01 per million tokens on short prompts and $0.05 on long ones, against $0.10 on Haiku 4.5.

Anthropic says Haiku 5.5 is priced 90% below Haiku 4.5 for requests up to 100,000 tokens and 50% below it for longer ones. About 90% of requests to the previous Haiku fell into the shorter tier. The company also says Haiku 5.5 uses an updated tokenizer, similar to those in Sonnet 5.5 and Opus 5.5, which consumes slightly more tokens per task. Its average-cost estimate accounts for that change.

A small model posting larger-model scores

Anthropic’s benchmark table puts Haiku 5.5 well ahead of its predecessor and, on several tests, ahead of OpenAI’s GPT-6 Luna:

• Computer use: 72.4% on the OSWorld 2.1 offline subset, against 15.7% for Haiku 4.5, 48.9% for GPT-6 Luna and 83.9% for Sonnet 5.5.
• Agentic coding: 39.2% on Terminal-Bench 4.0, up from 0.0% for Haiku 4.5, with GPT-6 Luna at 16.4% and Sonnet 5.5 at 70.6%. On FrontierCode 1.1 (Main), Haiku 5.5 scores 46.4% to GPT-6 Luna’s 42.4%.
• Knowledge work: 1,620 on GDPval-AA v2.1 and 1,578 on AA-Briefcase v1.1. Haiku 4.5 scored 735 and 614.
• Reasoning and charts: 45.9% on Humanity’s Last Exam without tools and 57.4% with them, plus 46.4% on the Chartography visual reasoning test.

These are Anthropic’s own results, and the company is clear about the model’s limits. Sonnet 5.5 and Opus 5.5 “remain better choices for complex agentic coding tasks,” Anthropic says. Haiku 5.5 is meant for narrowly scoped work that “might otherwise have been cost-prohibitive” on earlier Claude models.

Customer figures in the announcement point the same way. HubSpot says Haiku 5.5 scored 92.8%, averaged over three runs, on its internal CRM suite, the best score it has seen on that suite. AlphaSense reports a statistically significant gain over Haiku 4.5 on a 400-query document task, 0.84 against 0.76.

Effort control, context and safeguards

The effort setting lets developers trade intelligence for cost, as they already can on Anthropic’s larger models. The models overview in Anthropic’s documentation lists Haiku 5.5 with adaptive thinking, a default effort of medium, a 1-million-token context window, 128,000 maximum output tokens and a reliable knowledge cutoff of June 2026.

On safety, Anthropic reports “major improvements across almost all” of its alignment evaluations relative to Haiku 4.5. Its cyber safeguards are stricter than Haiku 4.5’s but somewhat looser than those on other recent Claude models. They allow a wider range of defensive tasks than Sonnet 5.5’s safeguards while still blocking penetration testing. Biology safeguards match Sonnet 5, Sonnet 5.5 and Opus 5.

Cheaper caching and API credits ride along

The launch came with wider price changes. Cache reads on Sonnet 5.5 drop by half, from $0.20 to $0.10 per million tokens, which Anthropic says makes the model around 20% cheaper on most agentic work. This week, Max 5x subscribers start receiving $100 a month in API credits, Max 20x subscribers $200, and Team plans up to $500 pooled across users. The Python and TypeScript SDKs also gain beta support for computer use and browser use.

Distribution started the same day. GitHub made Haiku 5.5 generally available in Copilot for Pro, Pro+, Max, Business and Enterprise users, with a gradual rollout. GitHub says that in early testing the model matched Claude Sonnet 5 on many coding tasks while using significantly fewer tokens and steps. For subagent and classification fleets, the short-prompt price cut may matter more than any benchmark.