By the toolist.ai editors. Researched from public sources, not hands-on tested by toolist.ai. Updated October 13, 2026; prices checked October 13, 2026.
Claude Haiku 5.5: price, context window and benchmarks
Claude Haiku 5.5 is Anthropic's small, fast model for high-volume work such as classification, extraction, routing and subagents, at $0.10 per million input tokens and $0.50 per million output tokens, a tenth of Claude Haiku 4.5's price. Prompts over 100,000 tokens cost five times as much ($0.50 in, $2.50 out), and Artificial Analysis scores it 43.4 on its Intelligence Index at max effort.
Best for
- High-volume, latency-sensitive jobs such as classification, extraction and routing
- Subagent, compaction and summary work under Claude Opus 5.5 or Sonnet 5.5; Anthropic still points complex agentic coding to the larger models
- Replacing Claude Haiku 4.5 on prompts up to 100,000 tokens, at a tenth of the token price
Key facts
| Fact | Claude Haiku 5.5 | Source |
|---|---|---|
| Provider | Anthropic | Claude docs: Pricing |
| API model name | claude-haiku-5-5 | Claude docs: Claude Haiku 5.5 |
| Released | October 7, 2026 | Anthropic: Introducing Claude Haiku 5.5 |
| Input price | $0.10 per million tokens | Claude docs: Pricing |
| Cached input price | $0.01 per million tokens | Claude docs: Pricing |
| Output price | $0.50 per million tokens | Claude docs: Pricing |
| Context window | 1,000,000 tokens | Claude docs: Claude Haiku 5.5 |
| Maximum output | 128,000 tokens | Claude docs: Claude Haiku 5.5 |
| Knowledge cutoff | June 2026 | Claude docs: Claude Haiku 5.5 |
| Input and output | Text and images in; text out | Claude docs: Claude Haiku 5.5 |
| Reasoning | Adaptive thinking is on by default and can be turned off at high effort or below. Effort levels low to max; medium is the API default | Claude docs: Effort |
| API and cloud platforms | Claude API, Claude Platform on AWS, Amazon Bedrock, Google Cloud and Microsoft Foundry | Claude docs: Claude Haiku 5.5 |
| In the apps | Haiku models are in the Claude apps on every plan, Free to Enterprise; the plan table names model families, not versions | Claude plans and pricing |
| Data retention | Available under zero data retention arrangements; it is not one of the models that require 30-day retention | Claude docs: API and data retention |
Prompts over 100,000 tokens are billed at $0.50 in, $2.50 out and $0.05 for cache reads per million tokens for the whole request; the prompt length counts cache reads and writes. Cache writes cost $0.125 per million tokens (5-minute cache) or $0.20 (1-hour cache), or $0.625 and $1 for prompts over 100,000 tokens. The Batch API costs half.
What Claude Haiku 5.5 costs a month
Three typical workloads, priced at the standard API rates above. We calculated these; the provider does not publish them.
| Workload | Tokens a month | Cost a month |
|---|---|---|
| Support chatbot 100,000 replies, each 3,000 input and 400 output tokens, no caching | 300M in, 40M out | $50.00 |
| Document Q&A 5,000 questions over the same 60,000-token document set (90% cache hits), 800 output tokens each | 300M in, 4M out | $7.70 |
| Coding agent 200 tasks, each 1.5 million input tokens (90% cache hits) and 50,000 output tokens | 300M in, 10M out | $10.70 |
Reasoning models bill their thinking as output tokens, so heavy reasoning can raise the output count several times. Cache-write charges are not included.
Benchmarks
| Benchmark | Score | Source |
|---|---|---|
| Artificial Analysis Intelligence Index max effort | 43.4 | Artificial Analysis: Claude Haiku 5.5 (max) |
| Intelligence Index at everyday effort medium effort, the API default | 34.5 | Artificial Analysis: Claude Haiku 5.5 (medium) |
| Coding Agent Index Claude Code harness, max effort | 36.5 | Artificial Analysis: Coding agents |
| Terminal-Bench 4.0 max effort | 32.8% | Artificial Analysis: Claude Haiku 5.5 (max) |
| Humanity's Last Exam max effort | 44.4% | Artificial Analysis: Claude Haiku 5.5 (max) |
| GDPval-AA v2.1 (Elo) max effort | 1620 | Artificial Analysis: Claude Haiku 5.5 (max) |
| SciCode max effort | 55.0% | Artificial Analysis: Claude Haiku 5.5 (max) |
| Terminal-Bench-Science 0.1 max effort | 20.0% | Artificial Analysis: Claude Haiku 5.5 (max) |
Speed: about 168 output tokens a second, 12.85 seconds to the first token (medium effort, Anthropic API; Artificial Analysis: Claude Haiku 5.5 (medium)).
Benchmark and speed data marked as such come from Artificial Analysis.
What changed
Compared with Claude Haiku 4.5, input and output prices fall from $1 and $5 to $0.10 and $0.50 per million tokens for prompts up to 100,000 tokens ($0.50 and $2.50 above that), and cache reads from $0.10 to $0.01 up to 100,000 tokens. It is the first Haiku with an adjustable effort setting, and its updated tokenizer uses slightly more tokens per task; Anthropic says it costs around 75% less to run on average. (Anthropic: Introducing Claude Haiku 5.5)
Claude Haiku 5.5 compared
Other models
- Gemini 4 Argon: $2 in, $10 out per million tokens
- Muse Wonder: $4 in, $20 out per million tokens
- GPT-6.1 Sol: $2 in, $10 out per million tokens
- Claude Sonnet 5.5: $2 in, $10 out per million tokens
- Claude Opus 5.5: $4 in, $20 out per million tokens
- GPT-6 Sol: $2 in, $10 out per million tokens
- GPT-6 Astra: $10 in, $50 out per million tokens
- Claude Fable 5.1: $10 in, $50 out per million tokens
- Claude Opus 5: $5 in, $25 out per million tokens
All prices side by side: AI model API price table. How we research: our method.
Sources
- Anthropic: Introducing Claude Haiku 5.5 (checked 2026-10-13)
- Claude docs: Claude Haiku 5.5 (checked 2026-10-13)
- Claude docs: What's new in Claude Haiku 5.5 (checked 2026-10-13)
- Claude docs: Effort (checked 2026-10-13)
- Claude docs: Pricing (checked 2026-10-13)
- Claude plans and pricing (checked 2026-10-13)
- Claude docs: API and data retention (checked 2026-10-13)
- Artificial Analysis: Claude Haiku 5.5 (max) (checked 2026-10-13)
- Artificial Analysis: Claude Haiku 5.5 (medium) (checked 2026-10-13)
- Artificial Analysis: Coding agents (checked 2026-10-13)
- OpenRouter: Claude Haiku 5.5 (checked 2026-10-13)