evals.report
BenchmarksLabsCompareRun guidesIn the wild
AnthropicClaude Haiku

Claude Haiku 4.5

Anthropic · Claude Haiku. Released Oct 15, 2025.

Claude Haiku 4.5 is a model from Anthropic in the Claude Haiku family, released Oct 15, 2025. evals.report tracks 30 reported Claude Haiku 4.5 benchmark scores across Berkeley Function Calling Leaderboard, DeepSWE, ARC-AGI-1, SWE-bench Pro, GPQA Diamond, FrontierMath, AIME (OTIS Mock), SimpleQA Verified, and 22 more — each shown with its benchmark, metric, source status, and date, and never combined into a single ranking.

30 results

Benchmark results 30

Compare this model
BenchmarkCategoryScoreMetricStatusDate
Berkeley Function Calling LeaderboardTool use68.70%accuracyOfficialOct 15, 2025Details
DeepSWECoding0.22%% resolvedOfficialOct 15, 2025Details
ARC-AGI-1Reasoning47.67%accuracyOfficialOct 15, 2025Details
SWE-bench ProCoding39.45%% resolvedOfficialOct 15, 2025Details
GPQA DiamondReasoning71.2%accuracyOfficialOct 15, 2025Details
FrontierMathReasoning5.9%accuracyOfficialOct 15, 2025Details
AIME (OTIS Mock)Reasoning66.7%accuracyOfficialOct 15, 2025Details
SimpleQA VerifiedOther5.9%accuracyOfficialOct 15, 2025Details
ARC-AGI-2Reasoning4.03%accuracyOfficialOct 15, 2025Details
WeirdMLCoding45.4%average accuracyOfficialOct 15, 2025Details
Epoch Capabilities IndexReasoning143.1IndexOfficialOct 15, 2025Details
MMLU-ProReasoning80.0%accuracyVerifiedOct 15, 2025Details
GAIA: A Benchmark for General AI AssistantsAgents56.36%accuracyUnverifiedOct 15, 2025Details
τ²-bench (Telecom)Tool use54.7%pass^1OfficialOct 15, 2025Details
GDPvalAgents1171EloOfficialOct 15, 2025Details
LiveCodeBenchCoding51.1%Pass@1UnverifiedOct 15, 2025Details
SciCodeCoding43.3%accuracyUnverifiedOct 15, 2025Details
MMMU (Massive Multi-discipline Multimodal Understanding and Reasoning Benchmark)Multimodal73.2%accuracyVerifiedOct 15, 2025Details
AA-Omniscience: Knowledge and Hallucination BenchmarkReasoning-4AA-Omniscience IndexOfficialOct 15, 2025Details
IFBenchReasoning54.3%accuracyOfficialOct 15, 2025Details
MultiChallengeReasoning50.49%accuracyVerifiedOct 15, 2025Details
Global-MMLUReasoning83.4%accuracyUnverifiedOct 15, 2025Details
WebDev ArenaChat preference1322EloVerifiedOct 15, 2025Details
Design ArenaChat preference1169EloVerifiedOct 15, 2025Details
MCP-UniverseTool use26.41%Overall Success RateVerifiedOct 15, 2025Details
SWE-bench MultilingualCoding64.7%% resolvedOfficialOct 15, 2025Details
FrontierMath Tier 4Reasoning2.1%accuracyOfficialOct 15, 2025Details
Vectara Hallucination LeaderboardOther9.8%Hallucination RateOfficialOct 15, 2025Details
Gray Swan Arena (Agent Red-Teaming / Indirect Prompt Injection)Agents1.3%Attack Success Rate (ASR)VerifiedOct 15, 2025Details
WebArenaAgents53.1%Task success rateVerifiedOct 15, 2025Details