Anthropic · The Agentic Era
Claude Haiku 4.5
The fast one. Used for sub-agents, high-volume classification, and the parts of an agent loop where capability matters less than throughput.
By C.W. Jameson · Published 19 May 2026 · Last reviewed 19 May 2026
Haiku 4.5 is what makes agent harnesses economically viable. When you need to summarise five thousand tool results, classify ten thousand emails, or run a search-result reranker every keystroke, Haiku does it for a fraction of Sonnet's cost and a quarter of its latency. It is not the model you give your hardest problem to. It is the model you give your hundredth-hardest problem to, ninety-nine times.
Field signature
Sub-second TTFT on short prompts. Will refuse less, but also reason less.
Specifications
| Released | 2025-10-01 |
|---|---|
| Context window | 200,000 tokens |
| Pricing | $1 / $5 per million tokens |
| Modalities | text · image |
| License | Commercial API only |
| Era | The Agentic Era |
Strengths
- Throughput
- Cost
- Low-latency tool routing
Weaknesses
- Will hallucinate more on out-of-distribution prompts
- Less consistent refusal behavior
Authentication markers
The fingerprints by which Claude Haiku 4.5 can be identified from its output alone.
| Tell | Meaning |
|---|---|
| Sub-300ms TTFT. | Haiku routing or a quantised open model. |
Notable works
- Default sub-agent in Claude Code
- Backbone of many search-rerank pipelines
Market position
$1-$5 per million tokens
Partner offer
Anthropic's Claude family is the model lineage most operators end up on for serious agent work. The free tier remains useful.
Try Claude →Affiliate link — see disclosure.
Primary sources
From the Almanac shop
The Operator's Compendium
Every agent harness, every routing pattern, every cost trick. 90-page PDF.
$29 — Coming soon