Data as of November 21, 2026 · How the score is built
Claude Haiku 5.5
Decision readingClaude Haiku 5.5 scores 63.5 out of 100 and ranks #38 of 216. This profile shows 39 source-displayable benchmark rows; its strongest eligible category is Agentic at #25. API pricing is $0.1 input and $0.5 output per million tokens, with cached input at $0.01.
Email me when its price, rank or availability changes.
Capability
63.5/100
field median 47.5#38 of 216 ranked models
Public
#38of 216
Verified #36 of 147
Price
$0.10input / $0.50 output
input median $0.60cached $0.010 · batch + cache $0.005 · blended $0.30
Speed
253tok/s
field median 93.5 tok/sFirst token 180.09 s
Context
1Mtokens
field median 256,000Maximum output length is tracked separately
Decision snapshot
Where this model sits in the ranked field, its strongest published evidence, and what to validate before choosing it. The marker compares it with the current ranked catalog, not with an absolute quality threshold.
Strongest published evidence
Agentic ranks #25. Particularly useful for coding agents, browser research, and computer-use workflows.
Validate before choosing
39 published rows leave some tracked benchmark slots empty. Knowledge is its lowest eligible category at #84.
Source-linked · 39 displayable benchmark rows
Is Claude Haiku 5.5 right for your job?
Its place on the LLM Selector’s job rankings, the same order the job pages publish. Describe your own job to see where it stands for that.
Where Claude Haiku 5.5 ranks by benchmark score:
We keep what you type here, drafts included, for 180 days, and after that only the job and requirements read from it. Nothing is tied to your IP address or account. Leave out names and other personal details. Privacy
Category score record
Scores and ranks appear only where published evidence can be displayed. The table keeps the score, weight, cohort, and evidence state together.
| Category | Score | Rank | Percentile | Weight | Benchmarks | Evidence |
|---|---|---|---|---|---|---|
| AgenticRank #25 of 182Percentile 87thWeight 22%9 benchmarksVerified | 68.3 | 9 benchmarks | Verified | |||
| CodingRank #28 of 180Percentile 85thWeight 20%7 benchmarksVerified | 67.9 | 7 benchmarks | Verified | |||
| ReasoningRank #38 of 161Percentile 77thWeight 17%4 benchmarksVerified | 62.6 | 4 benchmarks | Verified | |||
| MultimodalRank #39 of 119Percentile 68thWeight 12%4 benchmarksVerified | 63.9 | 4 benchmarks | Verified | |||
| KnowledgeRank #84 of 158Percentile 47thWeight 12%1 benchmarkVerified | 50.2 | 1 benchmark | Verified | |||
| MultilingualRank #36 of 88Percentile 60thWeight 7%Not listedUnlisted sources | 56.1 | Not listed | Unlisted sources | |||
| Inst. FollowingRank #37 of 143Percentile 75thWeight 5%Not listedUnlisted sources | 58.3 | Not listed | Unlisted sources | |||
| MathRank #29 of 125Percentile 77thWeight 5%2 benchmarksVerified | 62.1 | 2 benchmarks | Verified |
39 of 952 tracked benchmark slots have displayable evidence · bars run 0–100
Coverage detailsHow much of this is verified
Coverage is split by category so a strong number never hides a thin evidence base. Verified means the row is tied to a published source; provisional rows remain visible but separate.
- Agentic9/9 verified
- Coding7/7 verified
- Reasoning4/4 verified
- Multimodal4/4 verified
- Knowledge1/1 verified
- MultilingualNot measured
- Inst. FollowingNot measured
- Math2/2 verified
Capability shape
Each axis shows percentile within that category’s eligible cohort. The comparison outline is the median of the six nearest public-score peers; a collapsed vertex means the category is not rank-eligible.
Claude Haiku 5.5 category percentile values
- Agentic87th percentile
- Coding85th percentile
- Reasoning77th percentile
- Multimodal68th percentile
- Knowledge47th percentile
- Multilingual60th percentile
- Instruction following75th percentile
- Math77th percentile
The dashed outline is median of 6 nearest peers.
Eligible category ranks
- Agentic#25/182
- Coding#28/180
- Reasoning#38/161
- Multimodal#39/119
- Knowledge#84/158
- Multilingual#36/88
- Inst. Following#37/143
- Math#29/125
Benchmark ledger
Every result says who ran the test. For this model, Anthropic ran 7 of 40. Coding opens by default. The tick marks the best source-verified result in the catalog; provisional leaders do not set the reference.
Coding7 rows
| Benchmark | Score | Versus best verified row | Gap | Who ran it | Use in ranking |
|---|---|---|---|---|---|
| Bug Hunt Bench | 21.5 fixes | Best verified: Gemini 4 Neon · 63.0 fixes | Gap41.5 behind | Run by Bug Hunt Bench | Counts5% ref. weight |
| FrontierSWE v2 | 43.8% | Best verified: Claude Fable 5.2 · 69.9% | Gap26.1 behind | Run by FrontierSWE v2 | Counts4% ref. weight |
| VulcanBench Frontier v4 | 92.2% | Best verified: Claude Fable 5.2 · 93.8% | Gap1.7 behind | Run by VulcanBench Frontier v4 | Counts4% ref. weight |
| FrontierCode 1.1 Main | 46.4% | Best verified: Claude Opus 5.5 · 54.4% | Gap8 behind | Reported by Anthropic | Shown only |
| FrontierCode 2.0 | 28.9% | Best verified: Claude Fable 5.2 · 50.2% | Gap21.3 behind | Run by FrontierCode 2.0 | Shown only |
| CursorBench 4.0 | 48.4% | Best verified: Gemini 4 Carbon · 60.4% | Gap12 behind | Run by CursorBench 4.0 | Shown only |
| CursorBench 4.1 | 48.4% | Best verified: Gemini 4 Neon · 65.4% | Gap17 behind | Run by CursorBench 4.1 | Shown only |
Agentic9 rows
| Benchmark | Score | Versus best verified row | Gap | Who ran it | Use in ranking |
|---|---|---|---|---|---|
| Terminal-Bench 4.1 | 40.00% | Best verified: Claude Fable 5.2 · 65.80% | Gap25.8 behind | Run by Terminal-Bench 4.1 | Counts4% ref. weight |
| Gert LabsGert Labs Composite Game Benchmark | 58.62% | Best verified: GPT-6 Astra · 91.96% | Gap33.3 behind | Run by Gert Labs | Counts3% ref. weight |
| Terminal-Bench 4.0 | 39.20% | Best verified: Claude Sonnet 5.5 · 70.60% | Gap31.4 behind | Reported by Anthropic | Counts3% ref. weight |
| HLE w/ toolsHumanity's Last Exam with tools | 57.4% | Best verified: Kimi K3.1 Pro · 70.0% | Gap12.6 behind | Reported by Anthropic | Counts2% ref. weight |
| OSWorld 2.1 OfflineOSWorld 2.1 offline subset (82 tasks) | 72.4% | Best verified: Claude Fable 5.2 · 89.3% | Gap16.9 behind | Reported by Anthropic | Counts2% ref. weight |
| Agents' Last Exam (Linux)Agents' Last Exam, Linux subset | 57.0% | Best verified: Muse Ember · 61.6% | Gap4.6 behind | Lab-reported, from another publisher’s document | Shown only |
| MazeBenchMazeBench: visual spatial reasoning in a 3D open world | 1% | Best verified: GPT-6 Astra · 24% | Gap23 behind | Run by MazeBench | Shown only |
| Hermes IndexHermes Index: agentic capability and cost per task inside Hermes Agent | 37.0% | Best verified: Claude Opus 5.5 · 63.3% | Gap26.3 behind | Run by Hermes Index | Shown only |
| Hermes BenchHermes Bench: 150 agent tasks graded on the files and state the agent leaves behind | 71.0% | Best verified: Claude Opus 5.5 · 76.7% | Gap5.8 behind | Run by Hermes Bench | Shown only |
Reasoning4 rows
| Benchmark | Score | Versus best verified row | Gap | Who ran it | Use in ranking |
|---|---|---|---|---|---|
| ARC-AGI-2Abstraction and Reasoning Corpus for AGI v2 | 61.3% | Best verified: GPT-6.2 Astra · 95.8% | Gap34.6 behind | Run by ARC-AGI-2 | Counts2% ref. weight |
| ARC-AGI-1ARC-AGI-1 Semi-Private Evaluation | 89.00% | Best verified: Claude Fable 5 · 98.50% | Gap9.5 behind | Run by ARC-AGI-1 | Shown only |
| ARC-AGI-3Abstraction and Reasoning Corpus for AGI v3 | 0.5% | Best verified: GPT-6.2 Astra · 70.3% | Gap69.9 behind | Run by ARC-AGI-3 | Shown only |
| ARC-AGI-3 Provider AdapterARC-AGI-3 Semi-Private: Provider Adapter harness | 2.82% | Best verified: GPT-6.2 Astra · 99.80% | Gap97 behind | Run by ARC-AGI-3 Provider Adapter | Shown only |
Multimodal4 rows
| Benchmark | Score | Versus best verified row | Gap | Who ran it | Use in ranking |
|---|---|---|---|---|---|
| ParseBench-2ParseBench-2: severity-aware document parsing (LlamaIndex leaderboard) | 71.24% | Best verified: Claude Opus 5.5 · 87.11% | Gap15.9 behind | Run by ParseBench-2 | Counts5% ref. weight |
| ParseBench-1ParseBench-1: document parsing for AI agents (LlamaIndex leaderboard, first release) | 56.06% | Best verified: Claude Opus 5.5 · 78.80% | Gap22.7 behind | Run by ParseBench-1 | Counts5% ref. weight |
| PerceptionBench (tools)PerceptionBench with tools | 78.6% | Best verified: Muse Ember · 80.6% | Gap2 behind | Lab-reported, from another publisher’s document | Shown only |
| Chartography (no tools)Chartography without tools | 46.4% | Best verified: Claude Fable 5.2 · 73.4% | Gap27 behind | Reported by Anthropic | Shown only |
Knowledge1 row
| Benchmark | Score | Versus best verified row | Gap | Who ran it | Use in ranking |
|---|---|---|---|---|---|
| HLE w/o toolsHumanity's Last Exam without tools | 45.9% | Best verified: Claude Fable 5.2 · 73.2% | Gap27.3 behind | Reported by Anthropic | Counts4% ref. weight |
Math2 rows
| Benchmark | Score | Versus best verified row | Gap | Who ran it | Use in ranking |
|---|---|---|---|---|---|
| FrontierMath v2 (Tiers 1-3)FrontierMath v2 Tiers 1-3 | 75.088% | Best verified: Claude Fable 5.2 · 95.088% | Gap20 behind | Run by FrontierMath v2 (Tiers 1-3) | Counts11% ref. weight |
| FrontierMath v2 (Tier 4)FrontierMath v2 Tier 4 | 46.341% | Best verified: Claude Fable 5.2 · 100.000% | Gap53.7 behind | Run by FrontierMath v2 (Tier 4) | Counts2% ref. weight |
External signals12 rows
| Benchmark | Score | Versus best verified row | Gap | Who ran it | Use in ranking |
|---|---|---|---|---|---|
| Game Arena Go | 1076 | Best verified: Claude Fable 5.2 · 1877 | Gap801 behind | Run by Game Arena Go | Shown only |
| ASCII Arena | 1337 | Best verified: Gemini 4 Carbon · 2005 | Gap668 behind | Run by ASCII Arena | Shown only |
| Pelican Arena | 1305 | Best verified: Gemini 4 Neon · 1808 | Gap503 behind | Run by Pelican Arena | Shown only |
| FrontierPhysicsEnd-to-End Physics Research Benchmark for AI Agents | 4.2% | Best verified: Claude Fable 5.2 · 33.3% | Gap29.1 behind | Run by FrontierPhysics | Shown only |
| NYT ConnectionsNYT Connections LLM Benchmark (extended version) | 65.7% | Best verified: Gemini 4 Carbon · 99.3% | Gap33.6 behind | Run by NYT Connections | Shown only |
| FACTS Multimodal | 38.9% | Best verified: Gemini 4 Carbon · 60.2% | Gap21.3 behind | Run by FACTS Multimodal | Shown only |
| StealthMarkStealthMark Intelligence Benchmarks (overall score) | 49.5 | Best verified: GPT-6 Astra · 74.6 | Gap25.1 behind | Run by StealthMark | Shown only |
| Chess PuzzlesChess Puzzles (Epoch AI) | 55.0% | Best verified: Gemini 4 Neon · 83.0% | Gap28 behind | Run by Chess Puzzles | Shown only |
| Mystery Game PuzzlesMystery Game Puzzles (Epoch AI) | 30.0% | Best verified: GPT-6.2 Astra · 95.0% | Gap65 behind | Run by Mystery Game Puzzles | Shown only |
| Furniture AssemblyFurniture Assembly Benchmark (Epoch AI) | 47.5% | Best verified: Claude Fable 5.2 · 90.0% | Gap42.5 behind | Run by Furniture Assembly | Shown only |
| EBR-benchEarthborne Rangers (EBR-bench) | 43.8% | Best verified: GPT-6.2 Astra · 92.4% | Gap48.6 behind | Run by EBR-bench | Shown only |
| MirrorCodeMirrorCode (Epoch AI and METR) | 78.9% | Best verified: Claude Sonnet 5.5 · 97.8% | Gap18.9 behind | Run by MirrorCode | Shown only |
Other lab-reported results1 rowAnother metric, split or setting, or a test without a page
| Benchmark | Score | On a 0–100 scale | Who ran it | Use in ranking |
|---|---|---|---|---|
| FrontierCode 1.1 (Main)Task score · xhigh effort; the stored score is the max-effort run | 45.8% | Reported by Anthropic | Shown only |
Outlined bars are lab-reported. Filled bars are benchmark-run or independent.
What it costs to get this score
Published API price against the public score. The x-axis uses a log scale; the dashed path marks models that are not beaten by a cheaper, higher-scoring option. Price uses average of published input and output rates.
Claude Haiku 5.5 · 63.5 score · $0.30 blended per million tokens
The chart opens on the current model. Scroll horizontally to inspect the full price axis.
Horizontal: blended price per million tokens, log scale · Vertical: public score
Lineage
The sequence follows explicit supersedes links. Each score is estimated for that model; a relative can inform a sparse estimate but never sets a floor, so a newer release can score below an earlier one. Scores and prices remain blank when the corresponding public row or first-party rate is unavailable.
Base entry
Spec sheet
Each documented value carries its source. Missing fields stay visible as not sourced or not published, rather than disappearing from the page.
- API model ID
- claude-haiku-5-5Anthropic Claude Haiku 5.5 model documentation
- Context window
- 1MAnthropic Claude Haiku 5.5 model documentation
- Maximum output
- 128KAnthropic Claude Haiku 5.5 model documentation
- Knowledge cutoff
- June 2026Anthropic Claude Haiku 5.5 model documentation
- Input modalities
- text, imageAnthropic Claude Haiku 5.5 model documentation
- Output modalities
- textAnthropic Claude Haiku 5.5 model documentation
- Parameters
- Not disclosed by the provider
- Availability
- Claude API · Amazon Bedrock · Google Cloud · Microsoft Foundry · Claude Platform on AWSAnthropic Claude Haiku 5.5 model documentation
- Cloud regions
- Not tracked yet
- Lifecycle
- activeAnthropic Claude Haiku 5.5 model documentation
- API capabilities
- Tool calling, structured outputs, and batch support are not tracked yet
- Prompt caching
- Published at $0.010 per million cached input tokensClaude API pricing
- Self-host
- Weights are not published
- Rate limits
- Not tracked yet
How to read this profile
The visual layer above carries the decisions. These notes preserve the model, ranking, coverage, and family context behind the numbers.
Claude Haiku 5.5 ranks #38 of 216 on the public leaderboard with a score of 63.5/100. Its source-verified position is #36 of 147.
Claude Haiku 5.5 is a proprietary model with a 1M context window. It uses an explicit reasoning mode, which can improve complex problem solving while adding latency and token use.
Generally available across the Claude API, Amazon Bedrock, Google Cloud, Microsoft Foundry, and Claude Platform on AWS. The API model ID is `claude-haiku-5-5`; Anthropic lists a 1M-token context window and up to 128K output tokens.
Its explicit predecessor is Claude Haiku 4.5. 39 of 952 tracked benchmark slots currently have displayable evidence. Missing categories stay blank.
Its strongest eligible category is Agentic at #25, while its lowest eligible position is Knowledge at #84. particularly useful for coding agents, browser research, and computer-use workflows.
Last updated November 21, 2026. Runtime fields remain blank until a sourced snapshot exists.
Deployment options
Self-host and provider-specific paths stay separate from benchmark evidence so operating constraints are visible before a score becomes the whole decision.
Where to run Claude Haiku 5.5 by API
OpenRouter is the one host we track that publishes a per-token price for it. Prices read from each host’s own pages on November 16, 2026.
| Host | Input /1M | Output /1M | Cached /1M | Context |
|---|---|---|---|---|
| OpenRouter | $0.10 | $0.50 | $0.01 | 1M |
Questions
How does Claude Haiku 5.5 perform overall in AI benchmarks?
Claude Haiku 5.5 ranks #38 out of 216 models on the public BenchAlign leaderboard, with a score of 63.5/100. Its evidence status is Supported, and this profile shows 39 source-displayable benchmark rows. The label describes evidence depth, not a provider quality claim; inspect category rows before choosing a workload.
Is Claude Haiku 5.5 good for knowledge and understanding?
Claude Haiku 5.5 ranks #84 out of 158 eligible models for knowledge and understanding, with a public category score of 50.2/100. Higher-ranked alternatives are available for workloads where this category decides the choice. Check the underlying rows before treating the aggregate as a workload guarantee.
Is Claude Haiku 5.5 good for coding and programming?
Claude Haiku 5.5 ranks #28 out of 180 eligible models for coding and programming, with a public category score of 67.9/100. Higher-ranked alternatives are available for workloads where this category decides the choice. Check the underlying rows before treating the aggregate as a workload guarantee.
Is Claude Haiku 5.5 good for mathematics?
Claude Haiku 5.5 ranks #29 out of 125 eligible models for mathematics, with a public category score of 62.1/100. Higher-ranked alternatives are available for workloads where this category decides the choice. Check the underlying rows before treating the aggregate as a workload guarantee.
Is Claude Haiku 5.5 good for reasoning and logic?
Claude Haiku 5.5 ranks #38 out of 161 eligible models for reasoning and logic, with a public category score of 62.6/100. Higher-ranked alternatives are available for workloads where this category decides the choice. Check the underlying rows before treating the aggregate as a workload guarantee.
Is Claude Haiku 5.5 good for agentic tool use and computer tasks?
Claude Haiku 5.5 ranks #25 out of 182 eligible models for agentic tool use and computer tasks, with a public category score of 68.3/100. Higher-ranked alternatives are available for workloads where this category decides the choice. Check the underlying rows before treating the aggregate as a workload guarantee.
Is Claude Haiku 5.5 good for multimodal and grounded tasks?
Claude Haiku 5.5 ranks #39 out of 119 eligible models for multimodal and grounded tasks, with a public category score of 63.9/100. Higher-ranked alternatives are available for workloads where this category decides the choice. Check the underlying rows before treating the aggregate as a workload guarantee.
Does Claude Haiku 5.5 have full benchmark coverage on BenchLM?
No. Claude Haiku 5.5 currently has 64 source-displayable rows. The profile exposes published, non-generated evidence and leaves missing categories blank until an exact evaluation is available. Coverage describes how much was measured; it is not a penalty added to an individual benchmark result.
What is the context window size of Claude Haiku 5.5?
Claude Haiku 5.5 has a documented context window of 1M. That figure is the maximum combined prompt and retained-conversation space reported for this exact model; it is not the maximum output length. The profile keeps output limits separate because providers often publish those limits independently.