Anthropic released Claude Haiku 5.5 on October 7, 2026, at $0.10 per million input tokens and $0.50 per million output tokens for prompts up to 100,000 tokens. That is 90% below Claude Haiku 4.5's list price, and Anthropic says it costs around 75% less to run on average once a new tokenizer and different token use are counted. Above 100,000 prompt tokens, the whole request is billed at five times those rates.
Haiku 5.5 is the third model in the Claude 5.5 family, after Claude Opus 5.5 and Claude Sonnet 5.5. In its announcement, Anthropic aims it at high-volume, cost-sensitive work such as summaries, compaction, database queries and classification, and at subagent roles alongside Opus 5.5 and Sonnet 5.5. Anthropic calls it its fastest model at standard speed, while noting that Opus models in fast mode run quicker. It is also the first Haiku model with an adjustable effort setting.
Anthropic's models overview lists a 1M-token context window, 128K maximum output tokens, a June 2026 knowledge cutoff, a default effort of medium and retirement no sooner than October 7, 2027. The model ID is claude-haiku-5-5 on the Claude API, with no date suffix and no separate alias, and it is available on Amazon Web Services, Google Cloud and Microsoft Azure.
| Per million tokens (USD) | Prompt up to 100K | Prompt over 100K | Claude Haiku 4.5 |
|---|---|---|---|
| Input | $0.10 | $0.50 | $1.00 |
| Output | $0.50 | $2.50 | $5.00 |
| Cache writes (5-minute) | $0.125 | $0.625 | $1.25 |
| Cache writes (1-hour) | $0.20 | $1 | $2 |
| Cache reads | $0.01 | $0.05 | $0.10 |
Sources: Anthropic pricing documentation and announcement, checked October 10, 2026.
Anthropic's long context pricing rules say a request's prompt length counts every input token, including cache reads and cache writes. A request over 100,000 tokens pays the higher prices on the entire request, even if most of its prompt is a cache hit. That makes Haiku 5.5 the exception in Anthropic's current lineup: the same page says Claude 4.6 and later models other than Haiku 5.5 bill the full 1M-token window at standard rates.
Anthropic says about 90% of requests to Haiku 4.5 had prompts under 100,000 tokens, which is why it priced the model this way. For workloads in that range, the price drop is large. For long-document work that routinely crosses 100,000 tokens, the saving falls to 50% against Haiku 4.5, and Sonnet 5.5's flat pricing may be the better comparison.
The Haiku 5.5 migration guide says Haiku 5.5 uses the newer tokenizer shared with Claude 4.7 and later models, and that the same text produces approximately 30% more tokens than on Haiku 4.5, depending on content. Images can cost more too: Haiku 5.5 uses a high-resolution image tier, and the guide's example puts a 2,000 by 1,500 pixel image at about 2.5 times the visual tokens of Haiku 4.5.
Putting those together: a 90% lower price per token, applied to roughly 30% more tokens, still leaves a large saving for text under the threshold, which is consistent with Anthropic's average of about 75%. But any budget, max_tokens limit or context-length check calibrated on Haiku 4.5 token counts needs to be recounted with the Haiku 5.5 tokenizer. A max_tokens value that fit Haiku 4.5 output may now cut it off, and a prompt that sat just under 100,000 tokens on the old tokenizer may land above the new price threshold.
The migration guide also lists:
thinking: {"type": "enabled", "budget_tokens": N} returns a 400 error. Use adaptive thinking (the default) and set output_config.effort; lower effort where Haiku 4.5 ran without thinking. Responses can start with thinking blocks, so read content blocks by type.computer_toolset_20260801 toolset; a request that declares computer_20250124 returns a 400 error.refusal stop reason.model_context_window_exceeded stop reason. The guide does not list these for teams moving from Haiku 4.5.Anthropic's announcement compares Haiku 5.5 with Haiku 4.5, OpenAI's GPT-6 Luna and Sonnet 5.5. It reports large gains over Haiku 4.5, including 72.4% on the OSWorld 2.1 offline subset against 15.7%, and higher scores than GPT-6 Luna on the knowledge-work, computer-use and coding evaluations it lists. It also shows Haiku 5.5 well behind Sonnet 5.5 on Terminal-Bench 4.0 (39.2% against 70.6%) and says Sonnet 5.5 and Opus 5.5 remain the better choices for complex agentic coding.
These are Anthropic's figures, including the GPT-6 Luna scores, and this article does not reproduce them. The more durable point is Anthropic's own positioning: Haiku 5.5 is for narrowly scoped tasks that were too expensive to run at scale on earlier models.
Teams on Haiku 4.5 with short and medium prompts get the biggest saving. Recount tokens, revisit max_tokens and update thinking settings before switching.
Agent builders can use Haiku 5.5 for subagents, summarization and compaction steps under Sonnet 5.5 or Opus 5.5, keeping each subagent's prompt under 100,000 tokens.
Teams comparing low-cost tiers will note that GPT-6 Luna also lists at $0.10 input and $0.50 output per million tokens. Luna's long-prompt threshold is 272,000 tokens rather than 100,000, and it charges 2x input and 1.5x output above it, against 5x on both for Haiku 5.5. Identical headline prices hide very different long-context costs.
claude-haiku-5-5.max_tokens and context checks.budget_tokens thinking with adaptive thinking and an explicit effort level.computer_toolset_20260801 on the Claude API and Google Cloud, and keep conversations append-only.refusal stop reason; if you are coming from Haiku 3.5 or earlier, also update legacy tool versions and handle model_context_window_exceeded.When was Claude Haiku 5.5 released?
October 7, 2026.
How much does Claude Haiku 5.5 cost?
For prompts up to 100,000 tokens: $0.10 input, $0.50 output, $0.01 cache reads per million tokens. Over 100,000 tokens: $0.50 input, $2.50 output, $0.05 cache reads. Prices from Anthropic's pricing documentation, checked October 10, 2026.
Why does Claude Haiku 5.5 use more tokens than Haiku 4.5?
It uses a newer tokenizer that Anthropic says produces about 30% more tokens for the same text.
Discover more content: