Claude Haiku 5.5
Fast Claude model for responsive assistance, classification, and lightweight agents
Shogo Agent Index
Estimated#18 of 9257.5range 48.4–66.5 · 23% of weight measured
Agentic coding
67
26% measured
Long-horizon autonomy
62
30% measured
Knowledge-work agents
46
32% measured
Computer use
47
estimated
Security & context learning
55
estimated
Price
- Input
- $0.10/M
- Output
- $0.50/M
- Cache read
- $0.010/M
- 3:1 blend
- $0.20/M
- Agent-effective
- $0.050/M
Cheapest host for agents: Amazon Bedrock at $0.050/M.
Agent-effective = 20 input tokens per output token, 85% cache hits. Why.
Agentic benchmarks
| Benchmark | Score | Normalised | Index weight | Harness / source |
|---|---|---|---|---|
| SWE-bench Verified | est. | 8% | predicted 83 | |
| Terminal-Bench 2.0 | est. | 8% | predicted 86 | |
| DeepSWE | est. | 6% | predicted 64 | |
| FrontierSWE | 43.8% | 5% | proximus · $20.55/task | |
| CursorBench | 48.4% | 4% | source · $1.12/task | |
| GSO-Bench | est. | 4% | predicted 48 | |
| METR Time Horizons | est. | 10% | predicted 80 | |
| Vending-Bench 2 | $3,602 | 6% | source | |
| PostTrainBench | est. | 4% | predicted 30 | |
| APEX-Agents | 56.1% | 8% | source | |
| GDPval | est. | 7% | predicted 57 | |
| Remote Labor Index | est. | 5% | predicted 8 | |
| DeepResearch Bench | est. | 5% | predicted 50 | |
| OSWorld 2.0 | est. | 6% | predicted 19 | |
| OSWorld | est. | 4% | predicted 87 | |
| Cybench | est. | 5% | predicted 85 | |
| CL-bench | est. | 5% | predicted 25 |
Other benchmarks
| Benchmark | Category | Score | Source |
|---|---|---|---|
| ARC-AGI | Reasoning | 89.0% | source |
| ARC-AGI-2 | Reasoning | 61.3% | source |
| Chess Puzzles | Reasoning | 55.0% | source |
| CritPt | Reasoning | 22.6% | source |
| DTBench | Knowledge | 95.0% | source |
| FrontierCode | Coding | 46.4% | claude-code |
| FrontierMath Tiers 1-3 | Math | 75.1% | source |
| FrontierMath Tier 4 | Math | 46.3% | source |
| GPQA Diamond | Reasoning | 89.6% | source |
| Humanity's Last Exam | Reasoning | 30.1% | source |
| LMCA | Knowledge | 51.1% | source |
| MirrorCode | Coding | 78.9% | source |
| OTIS Mock AIME | Math | 98.9% | source |
| ProofBench | Math | 89.0% | source |
| SciCode | Coding | 55.0% | source |
| SimpleBench | Reasoning | 59.2% | source |
| SimpleQA Verified | Knowledge | 23.8% | source |
| WeirdML v3 | Coding | 11.7% | claude_code |
API hosts
Live prices and uptime for every host serving Claude Haiku 5.5 via OpenRouter.
| Host | Input | Output | Cache read | Agent $/M | Uptime 24h | Quant |
|---|---|---|---|---|---|---|
| Amazon Bedrock | $0.10 | $0.50 | $0.010 | $0.050 | 100.0% | — |
| Anthropic | $0.10 | $0.50 | $0.010 | $0.050 | 100.0% | — |
| Azure | $0.10 | $0.50 | $0.010 | $0.050 | 100.0% | — |
| Claude Platform on AWS | $0.10 | $0.50 | $0.010 | $0.050 | 100.0% | — |
| $0.10 | $0.50 | $0.010 | $0.050 | 100.0% | — |