Inference
EN
Anthropic

Claude Haiku 4.5

Claude Haiku 4.5 is a standard model from Anthropic. 19 benchmarks count toward its index, in 7 categories.

  • Closed weights
  • Standard
  • Input: text · image · file
  • Released 2025-10
RankedEnough results for an index.
IndexAll categories, out of 100.43.2±4.2Median 51.0
CoverageIndex weight that has results.95%
SpeedOutput tokens per second.88/sMedian 67/s
First tokenSeconds to the first output token.0.8 sMedian 1.3 s
Input / 1MUS dollars per 1M input tokens.$1.00batch $0.50Median $0.60
Output / 1MUS dollars per 1M output tokens.$5.00batch $2.50 Batch price per 1M output tokens.Median $2.55
ContextMaximum tokens in one request.200KMedian 262K
EloLMArena rating and rank.1397±2 #164Median 1422

Details

From OpenRouter
API IDThe model ID on OpenRouter.
anthropic/claude-haiku-4.5
Max outputMaximum output tokens per request.
64K
Knowledge cutoffLast date of training data.
—
ToolsTool calls on OpenRouter.
Yes
JSON outputOutput follows a given JSON schema.
Yes
Cache read / 1MUS dollars per 1M cached tokens.
$0.10
WeightsLink to the published weights.
—
RetiresDate OpenRouter removes the model.
—

Where to run

APIs that serve this model.
6 providers
ProviderTypeMaker, cloud, host or router.InputOutputContext
AnthropicMaker$1.00$5.00200K
DatabricksCloud$1.00$5.00200K
DigitalOceanCloud$1.00$5.00200K
OpenRouterRouter$1.00$5.00200K
SAP AI CoreCloud$1.00$5.00200K
Snowflake CortexCloud$1.00$5.00200K

US dollars per 1M tokens, list price. From models.dev (opens in a new tab) (MIT).

Capabilities

Index per category, out of 100.
Out of 100
AgenticMulti-step tasks with tools.4 counted48.9135 / 218
CodingCode writing and repair.3 counted49.9138 / 209
ReasoningLogic problems and puzzles.4 counted38.1176 / 227
MultimodalTasks with images and text.1 counted42.6—
KnowledgeFacts and expert knowledge.2 counted36.1217 / 268
MultilingualTasks in many languages.1 counted44.5—
InstructionTasks with strict prompt rules.None counted——
MathMath problems.4 counted40.5125 / 168

Results

19 counted
BenchmarkThe benchmark name.ResultThe score from the source.PlacePlace among results on this benchmark.
MATH level 5Math91.6% ±0.514 / 88
EuroEval SpanishMultilingual49.7% ±0.444 / 176
EuroEval GermanMultilingual47.8% ±0.648 / 191
SWE-bench Verified · SWE-bench/experimentsCoding66.6%18 / 65
EuroEval FrenchMultilingual57.6% ±0.452 / 178
EuroEval DutchMultilingual53.4% ±0.552 / 175
EuroEval ItalianMultilingual53.1% ±0.553 / 174
EuroEval SwedishMultilingual53.7% ±0.560 / 195
EuroEval PortugueseMultilingual53.4% ±0.658 / 182
EuroEval PolishMultilingual49.5% ±0.670 / 209
MazeBench: visual spatial reasoning in a 3D open worldAgentic0.0%23 / 48
Software Engineering Benchmark VerifiedCoding73.3%55 / 114
GPQA diamond · Epoch AIKnowledge65.8% ±3.2120 / 217
OTIS Mock AIME 2024-2025Math51.3% ±7.1121 / 200
FrontierMath-Tier-4-2025-07-01-PrivateMath2.1% ±2.135 / 55
FrontierMath-2025-02-28-PrivateMath5.0% ±1.445 / 69
LiveBench Browser UseAgentic40.0%34 / 49
ARC-AGI-2 (semi-private)Reasoning1.3%103 / 131
ParseBench: document parsing for AI agents (LlamaIndex leaderboard)Multimodal49.3%34 / 39
JobBenchAgentic16.0%32 / 36
ARC-AGI-1 (semi-private)Reasoning14.3%112 / 126
LiveBench DesignAgenticAbove fitFar above this model's index.59.9%48 / 52
SimpleQA VerifiedKnowledge12.9% ±1.084 / 91
LiveBench Data AnalysisReasoning59.3%74 / 80
VulcanBench v3Coding76.2%13 / 14
Chess PuzzlesReasoning6.5% ±2.784 / 90
LiveBench Coding 2.0Coding48.1%50 / 52
LiveBench ResearchAgentic42.2%50 / 52
LiveBench ReasoningReasoning61.7%78 / 80
LiveBench Long AgenticAgentic11.1%50 / 51
OSWorld 2.1 offline subset (82 tasks)Agentic15.7%10 / 10
FrontierMath-2025-02-28-PublicMath0.0%33 / 64
CorpFin v2Finance60.3% ±1.069 / 131
ARC-AGI-2 (public eval)Reasoning0.0%102 / 122
Public Benefits Bench v1.1Public benefits54.3% ±1.342 / 50
SWE-bench Multilingual · SWE-bench/experimentsCoding64.7%12 / 14
ARC-AGI-1 (public eval)Reasoning26.6%101 / 116
Public Benefits Bench v1Public benefits49.5% ±1.312 / 13

Sources

BenchLMCC BY-NC 4.0OpenRouterTerms restrict reuseEpoch AICC BYSWE-bench teamNo licence statedLiveBenchNo licence statedEuroEvalMITARC Prize FoundationTerms restrict reusemodels.devMIT

History

5 releases
Index44.0 → 43.2
23 Oct 202612 Nov 2026
Output / 1M$5.00 → $5.00
23 Oct 202612 Nov 2026

Questions

How much does Claude Haiku 4.5 cost?
$1.00 per 1M input tokens, $5.00 per 1M output tokens.
How good is Claude Haiku 4.5?
Its index is 43.2 out of 100. That is place 191 of 283 ranked models.
What is Claude Haiku 4.5 best at?
Agentic: 48.9 out of 100, place 135 of 218.
How fast is Claude Haiku 4.5?
88 output tokens per second, 0.8 s to the first token.
What is the context window of Claude Haiku 4.5?
200K tokens.
Is Claude Haiku 4.5 open source?
No. Its weights are not published.
When was Claude Haiku 4.5 released?
October 2025.

Badge

Claude Haiku 4.5 on Inference360: place 191 of 283, index 43.2
HTML
<a href="https://inference360.ai/models/claude-haiku-4-5"><img src="https://inference360.ai/badges/claude-haiku-4-5.svg" alt="Claude Haiku 4.5 on Inference360: place 191 of 283, index 43.2" height="20"></a>
Markdown
[![Claude Haiku 4.5 on Inference360: place 191 of 283, index 43.2](https://inference360.ai/badges/claude-haiku-4-5.svg)](https://inference360.ai/models/claude-haiku-4-5)

Compare with

GPT-OSS 120B43.4Claude Sonnet 443.0Claude Opus 4.5 Thinking56.3

Lower price

dots3-note Preview63.4 · FreeApodex 1.1 Mini55.7 · FreeNemotron 3 Ultra50.8 · Free

More from Anthropic

All models
Claude Haiku 4.5 Thinking43.5Claude Sonnet 4.5 Thinking53.8Claude Sonnet 4.547.9Claude Opus 4.5 Thinking56.3Claude Opus 4.552.0