Anthropic's fast lightweight language model for classification, extraction, routing, and high-volume subagents, with million-token context, vision, and adaptive thinking.
Anthropic
| Item | Specification |
|---|---|
| Input | Text, images |
| Output | Text |
| Context | 1,000,000 tokens |
| Max Output | 128,000 tokens; up to 300,000 tokens in Message Batches API beta |
| Vision | ✓ Supported |
| Function Calling | ✓ Supported |
Claude Haiku 5.5 is a fast language model released by Anthropic on October 7, 2026. Its official Claude API identifier is claude-haiku-5-5. It accepts text and images, produces text, provides a one-million-token context window with up to 128,000 standard output tokens, and has a reliable knowledge cutoff of June 2026.
The model targets latency-sensitive classification, routing, extraction, and high-volume subagent work. Anthropic marks it as Fastest in comparative latency across the current Claude lineup. Adaptive Thinking is on by default with medium as the default effort, while vision and tool use make it suitable for well-scoped execution inside larger agent systems.
effort to adjust depth.| Feature | Benefit |
|---|---|
| Fastest Standard-Speed Claude | Fits real-time experiences, high-throughput processing, and automation requiring frequent feedback. |
| 1M Context Window | Handles large source sets, long conversations, and broad project context within a lightweight model. |
| 128K Standard Output | Produces substantial batch results, code, analysis, and structured documents. |
| Adaptive Thinking | Keeps routine tasks direct and uses effort to add depth for harder problems. |
| Vision and Tool Coordination | Reads an interface, screenshot, or document before invoking tools to continue processing. |
| Subagent Specialization | Works well under Sonnet, Opus, or another orchestrator for many independent, well-scoped tasks. |
effort for Claude Haiku 5.5?The default medium level is a useful starting point for most evaluations. Test lower levels for straightforward classification, extraction, and routing, and raise effort for subtasks involving multistep judgment or tools. Compare accuracy, completion, and observed latency on a fixed sample set.
Anthropic requires applications to omit temperature, top_p, and top_k. A non-default value for any of them returns a 400 error, so control behavior with clear prompts, output schemas, and effort instead of carrying over older sampling configurations.
Context increases from 200K to one million tokens and standard output rises from 64K to 128K, but the new tokenizer generally produces more tokens for the same text. Read replies by content-block type, keep conversation history append-only when replaying thinking blocks, and recount existing prompts.
Start with Haiku 5.5 for classification, extraction, routing, frequent automation, and clearly decomposed subagent work. Choose Sonnet 5.5 for stronger everyday coding, agent orchestration, complex judgment, and greater multistep stability, then compare both on quality, tool success, and latency.
Claude Haiku 5.5 is available on LinkModel. Open its live model page, copy the published model ID, and submit text or image input with the current request structure. Use the live page for the platform route, context limits, tool support, and thinking parameters.
Complete a basic text request with the ID shown on LinkModel, then test classification, extraction, batch routing, image input, long context, and tool calls. Record response fields, thinking blocks, truncation, errors, and concurrent behavior, and confirm which options LinkModel currently exposes for effort, Batch, and other Claude extensions.
Tiered by input prompt tokens (incl. cache): once over the threshold, the whole request is billed at the higher tier.
| Token Type | Short context ≤100K | Long context >100K |
|---|---|---|
| Input | $0.085$0.1-15% | $0.425$0.5-15% |
| Output | $0.425$0.5-15% | $2.125$2.5-15% |
| Cached input | $0.0085$0.01-15% | $0.0425$0.05-15% |
| Cache write (5 minutes) | $0.10625$0.125-15% | $0.53125$0.625-15% |
| Cache write (1 hour) | $0.17$0.2-15% | $0.85$1-15% |