AnthropicClaude Haiku 5.5
Claude Haiku 5.5 is an Anthropic proprietary reasoning model, released 7 Oct 2026. Its best configuration (high reasoning effort) ranks #92 of 760 on the BenchLeader Index at 61.3 ±3.2, in the upper half. The ± is the point: 176 other configurations score within that range, so they and this one cannot be told apart on quality alone — price and speed are what separate them. It scores highest in composite (72) and lowest in coding (63). At $0.200 per million tokens blended it is cheaper than most ranked models. Output speed of 174 tokens per second puts it in the fastest quarter, with a first answer in 22.6 s. It has been measured at 6 reasoning-effort settings; this summary describes the best-scoring one, and the tabs above switch between them. Last measured 8 Oct 2026.
- Blended price
- $0.200/M
- $0.100 in · $0.500 out
- Output speed
- 174 tok/s
- measured by Artificial Analysis
- First answer
- 23 s
- first token 1.75 s
- Context
- 1M
- Full answer
- 26 s
- median, reasoning included
- Cost per run
- $0.079
- one full Intelligence Index run
- Released
- 7 Oct 2026
- Overall index61
- Reasoning72
- Coding63
- Maths66
- Knowledge64
- Long context64
- Composite72
Versions
Anthropic has shipped 6 models under this name. Each is ranked on its own results; a newer version often has fewer results so far, which holds its index nearer the average until more arrive.
| Model | Released | Index | Rank |
|---|---|---|---|
| NewClaude Haiku 5.5highthis page | 7 Oct 2026 | 61.3 | #92 |
| Claude Haiku 4.5no reasoning | 15 Oct 2025 | 48.5 | #409 |
| Claude 3.5 Haiku | 4 Nov 2024 | 39.8 | #650 |
| Claude 3 Haiku | 7 Mar 2024 | 37.1 | #714 |
| Claude Haiku 3 | – | – | – |
| Claude Haiku 4thinking | – | – | – |
Reasoning-effort configurations
The same model behaves differently depending on how much it is allowed to think. Each row is one setting, scored only on the benchmarks that were run at that setting. “Not stated” collects results from publishers that did not say which setting they used; for a reasoning model that is usually its thinking mode, but we do not assume it. Pick a setting here or at the top of the page to see its price, speed and category scores.
| Effort | Index | Rank | Speed | First answer | Chat reply cost | Categories |
|---|---|---|---|---|---|---|
| low | – | – | 178 tok/s | 9.72 s | $0.0002 | Composite 63 · Knowledge 63 · Long context 61 · Reasoning 56 |
| medium | – | – | 166 tok/s | 13 s | $0.0002 | Composite 69 · Knowledge 64 · Long context 64 · Reasoning 63 |
| highbest | 61.3 | #92 | 174 tok/s | 23 s | $0.0002 | Coding 63 · Composite 72 · Knowledge 64 · Long context 64 · Maths 66 · Reasoning 72 |
| xhigh | 61.1 | #94 | 188 tok/s | 57 s | $0.0002 | Composite 59 · Knowledge 64 · Long context 64 · Maths 67 · Reasoning 79 |
| max | 57.9 | #157 | 240 tok/s | 289 s | $0.0002 | Composite 54 · Knowledge 51 · Long context 67 · Maths 62 · Reasoning 69 |
| not stated | – | – | 146 tok/s | 1.75 s | $0.0002 | Instruction following 64 |
What more thinking costs
Turning the effort up buys index points and multiplies the bill. Here max costs 2.7× high for -3.4 on the index — -2.4 points per doubling of spend. Compare that with other models.
Benchmark results
One column per reasoning effort. Rank is among every configuration of every model on that benchmark. Hover a score for the run it came from.
Reasoning
| Benchmark | low | medium | high | xhigh | max | not stated | Source | Trend |
|---|---|---|---|---|---|---|---|---|
| GPQA Diamond | – | – | – | – | 89.7%#50 | – | Epoch AI Benchmarking Hub | |
| LiveBench Reasoningnot in index | – | – | – | 81.2%#49 | 84.6%#39 | – | LiveBench | |
| Humanity's Last Exam (AA)not in index | 27.0%#166 | 33.8%#121 | 37.3%#99 | 42.7%#58 | 44.4%#48 | – | Artificial Analysis | |
| CritPt | 9.1%#118 | 12.9%#97 | 18.6%#63 | 22.6%#51 | 18.9%#62 | – | Artificial Analysis | |
| Mystery Game Puzzlesnot in index | – | – | – | – | 30.0%#31 | – | Epoch AI Benchmarking Hub |
Coding
| Benchmark | low | medium | high | xhigh | max | not stated | Source | Trend |
|---|---|---|---|---|---|---|---|---|
| LMArena WebDev | – | – | 1587#29 | – | – | – | LMArena | |
| LiveBench Codingnot in index | – | – | – | 76.4%#45 | 77.9%#34 | – | LiveBench | |
| SciCode (AA)not in index | 49.2%#111 | 49.0%#114 | 48.7%#115 | 51.7%#85 | 55.0%#50 | – | Artificial Analysis | |
| Terminal-Bench 4.0 (AA)not in index | 12.6%#73 | 15.2%#65 | 21.7%#54 | 29.3%#44 | 32.8%#38 | – | Artificial Analysis |
Agents & tools
| Benchmark | low | medium | high | xhigh | max | not stated | Source | Trend |
|---|---|---|---|---|---|---|---|---|
| LiveBench Agentic Codingnot in index | – | – | – | 51.4%#39 | 57.2%#21 | – | LiveBench | |
| GDPval-AA v2.1not in index | 31.1%#129 | 38.7%#99 | 45.9%#68 | 50.5%#43 | 55.9%#24 | – | Artificial Analysis | |
| AutomationBenchnot in index | 22.9%#45 | 28.6%#44 | 33.7%#43 | 36.0%#40 | 35.4%#42 | – | Artificial Analysis | |
| GDP.pdfnot in index | 11.2%#43 | 15.2%#39 | 17.2%#34 | 18.2%#32 | 20.8%#25 | – | Artificial Analysis | |
| Harvey LABnot in index | 1.7%#13 | 0.6%#17 | 1.1%#15 | 2.8%#11 | 0.6%#17 | – | Harvey | |
| AA-Briefcase v1.1not in index | 1112#43 | 1372#38 | 1442#32 | 1533#21 | 1577#13 | – | Artificial Analysis |
Maths
| Benchmark | low | medium | high | xhigh | max | not stated | Source | Trend |
|---|---|---|---|---|---|---|---|---|
| FrontierMath Tiers 1–3 | – | – | – | – | 75.1%#18 | – | Epoch AI Benchmarking Hub | |
| FrontierMath Tier 4 | – | – | – | – | 46.3%#25 | – | Epoch AI Benchmarking Hub | |
| OTIS Mock AIME | – | – | 97.2%#32 | 98.9%#16 | – | – | Epoch AI Benchmarking Hub | |
| LiveBench Mathematicsnot in index | – | – | – | 92.8%#26 | 86.4%#49 | – | LiveBench |
Knowledge
| Benchmark | low | medium | high | xhigh | max | not stated | Source | Trend |
|---|---|---|---|---|---|---|---|---|
| SimpleQA Verified | – | – | – | – | 23.8%#67 | – | Epoch AI Benchmarking Hub | |
| LiveBench Data Analysisnot in index | – | – | – | 72.8%#43 | 51.8%#66 | – | LiveBench | |
| AA-Omniscience | 3.3#110 | 4.5#105 | 5.8#99 | 6.0#97 | 10.7#89 | – | Artificial Analysis | |
| AA-Omniscience: accuracynot in index | 33.3%#156 | 34.0%#152 | 34.8%#149 | 35.0%#145 | 36.4%#142 | – | Artificial Analysis | |
| AA-Omniscience: non-hallucinationnot in index | 55.1%#93 | 55.2%#91 | 55.5%#89 | 55.6%#88 | 59.6%#74 | – | Artificial Analysis |
Instruction following
| Benchmark | low | medium | high | xhigh | max | not stated | Source | Trend |
|---|---|---|---|---|---|---|---|---|
| LiveBench Languagenot in index | – | – | – | 63.5%#62 | 62.6%#64 | – | LiveBench | |
| LiveBench Instruction Followingnot in index | – | – | – | 66.5%#39 | 54.7%#64 | – | LiveBench | |
| MultiChallengeeffort not stated | – | – | – | – | – | 67.7%#6 | Scale AI SEAL |
Long context
| Benchmark | low | medium | high | xhigh | max | not stated | Source | Trend |
|---|---|---|---|---|---|---|---|---|
| AA-LCR | 72.0%#183 | 77.3%#124 | 77.3%#124 | 78.3%#106 | 82.7%#33 | – | Artificial Analysis | |
| MLCRnot in index | – | – | – | – | 63.3%#5 | – | Artificial Analysis |
Composite
| Benchmark | low | medium | high | xhigh | max | not stated | Source | Trend |
|---|---|---|---|---|---|---|---|---|
| LiveBench | – | – | – | 72.1%#49 | 67.9%#58 | – | LiveBench | |
| AA Intelligence Index v4.3.2 | 29.4#120 | 34.5#86 | 37.8#76 | 41.3#53 | 43.4#43 | – | Artificial Analysis |
Safety & honesty
| Benchmark | low | medium | high | xhigh | max | not stated | Source | Trend |
|---|---|---|---|---|---|---|---|---|
| FORTRESSnot in indexeffort not stated | – | – | – | – | – | 9.5%#66 | Scale AI SEAL |
Where to run it
Every provider serving this model through OpenRouter, with throughput and first-token latency measured on live traffic over the last 30 minutes and each provider’s own price. Purple marks the best in each column.
| Provider | Speed | First token | Input $/M | Output $/M | Context | Quantisation |
|---|---|---|---|---|---|---|
| Google Vertex | 218 tok/s | 0.72 s | $0.110 | $0.550 | 1M | – |
| Google Vertex | 173 tok/s | 1.48 s | $0.110 | $0.550 | 1M | – |
| Anthropic | 167 tok/s | 1.69 s | $0.100 | $0.500 | 1M | – |
| Google Vertex | 156 tok/s | 1.43 s | $0.100 | $0.500 | 1M | – |
| Amazon Bedrock | 149 tok/s | 2.70 s | $0.100 | $0.500 | 1M | – |
| Claude Platform on AWS | 143 tok/s | 1.94 s | $0.100 | $0.500 | 1M | – |
| Azure | 124 tok/s | 0.74 s | $0.100 | $0.500 | 1M | – |
| Amazon Bedrock | 74 tok/s | 0.82 s | $0.110 | $0.550 | 1M | – |
| Amazon Bedrock | – | 0.60 s | $0.110 | $0.550 | 1M | – |
What a task costs
Estimates from list price, output speed and time to first answer for the best configuration. “With caching” assumes three-quarters of the input is served from the prompt cache at $0.010 per 1M. Reasoning tokens are not modelled.
| Workload | Tokens in / out | Cost | With caching | Time |
|---|---|---|---|---|
| Chat reply | 400 / 300 | $0.0002 | $0.0002 | 24.4 s |
| Summarise a 30-page report | 12,000 / 600 | $0.0015 | $0.0007 | 26.1 s |
| Code edit | 6,000 / 1,500 | $0.0014 | $0.0009 | 31.3 s |
| Agentic coding session | 60,000 / 4,000 | $0.0080 | $0.0040 | 45.6 s |
| Structured extraction | 2,000 / 200 | $0.0003 | $0.0002 | 23.8 s |
See also
Data as of 11 Oct 2026. Compare these configurations.
Cite as: BenchLeader, “Claude Haiku 5.5: benchmarks, pricing, speed and rank”, https://www.benchleader.com/models/claude-haiku-5-5, data as of 11 Oct 2026.