Claude Haiku 5.5

What Anthropic says. The choosing guide describes an "efficiency-first" approach: begin with Haiku 5.5, test thoroughly, and upgrade only if needed for specific capability gaps. It recommends this route for prototyping, tight latency requirements, cost-sensitive implementations and high-volume, straightforward tasks. Its selection matrix lists real-time applications, high-volume intelligent processing, cost-sensitive deployments needing strong reasoning and sub-agent tasks. The guide also says to start Haiku 5.5 at its default medium effort and adjust from there.

Our editorial judgment. Haiku 5.5 is the model to try first for pipelines that make many short calls: tagging support tickets, extracting fields from forms, routing requests to a bigger model, or running as a worker under an orchestrator. Keep prompts under the 100,000 token threshold where you can, because the long-prompt price applies to the whole request. For open-ended drafting or multi-step coding, test Claude Sonnet 5.5 alongside it before committing. Packaged assistants are covered in our AI chatbots guide.

• Long-prompt pricing. Anthropic's long context pricing section says Haiku 5.5 is the exception to flat 1M context pricing. A request whose prompt is over 100,000 tokens pays the higher prices, and prompt length counts all input tokens, including cache reads and cache writes. A request over the threshold pays the higher prices even when part of its prompt is a cache hit.
• Tokenizer change. The model page says the same text counts as approximately 30% more tokens than on Claude Haiku 4.5, which also pushes prompts toward the 100,000 token line sooner.
• Thinking blocks are account-bound. They work only in the account that produced them, or in an account linked to it.
• Sampling parameters. Anthropic says to omit temperature, top_p and top_k; its migration guide lists the values that return a 400 error.
• Not usable as an advisor. The Sonnet 5.5 page says the advisor tool rejects Haiku 5.5 as an advisor model.
• US-only inference costs more. Setting inference_geo to "us" applies a 1.1x multiplier, and on Haiku 5.5 it also applies to the higher long-prompt prices.
• No fast mode. The pricing page lists fast mode only for Opus models.

All four current models share a 1M token context window, 128K max output, text and image input and a June 2026 reliable knowledge cutoff, according to the models overview.

The pricing page lists Claude Haiku 4.5 at USD 1 input and USD 5 output per 1M tokens, ten times Haiku 5.5's base rate. The deprecations page lists Haiku 4.5 as Active with a tentative retirement date of not sooner than 15 October 2026, so teams still on it should plan a migration test.

1. Pull 100 or more real inputs from the high-volume task you want to automate, with labels or expected outputs.
2. Run them on Haiku 5.5 at medium effort and on Sonnet 5.5 at high, with identical prompts, and measure accuracy against your labels.
3. Log prompt length per request and flag any that cross 100,000 tokens, since those are billed at the higher rate.
4. Measure response time at your expected concurrency, keep Haiku 5.5 where accuracy meets your bar, and route failures to a larger sibling.

We read Anthropic's models overview, Haiku 5.5 model page, pricing page, choosing a model guide and deprecations page on 15 October 2026. We have not run Claude Haiku 5.5, and nothing here is a test result or ranking. Check the vendor pages before committing budget. See our ranking method for how HumanRank scores tools and the model directory for other profiles.

What is Claude Haiku 5.5? It is Anthropic's fastest and lowest-priced current model, built for high-volume, latency-sensitive tasks such as classification, extraction, routing and subagent work.

How much does Claude Haiku 5.5 cost? It costs USD 0.10 per 1M input tokens and USD 0.50 per 1M output tokens for prompts up to 100,000 tokens, and USD 0.50 and USD 2.50 above that, at list price on 15 October 2026.

What is the Claude Haiku 5.5 context window? It is 1M tokens, with up to 128K output tokens, but prompts over 100,000 tokens are billed at higher rates.

Does Claude Haiku 5.5 support thinking? Yes, adaptive thinking is on by default, and you steer its depth with the effort parameter, which defaults to medium.

Is Claude Haiku 5.5 cheaper than Claude Haiku 4.5? Yes at list price for prompts up to 100,000 tokens, though the newer tokenizer produces about 30% more tokens for the same text.

When should I move from Haiku 5.5 to Sonnet 5.5? Move when your own accuracy tests show Haiku 5.5 missing your bar on a task type, which is the upgrade path Anthropic's guide describes.