evals.report
BenchmarksLabsCompareRun guidesIn the wild
AnthropicClaude Sonnet

Claude Sonnet 4.5

Anthropic · Claude Sonnet. Released Sep 29, 2025.

Claude Sonnet 4.5 is a model from Anthropic in the Claude Sonnet family, released Sep 29, 2025. evals.report tracks 46 reported Claude Sonnet 4.5 benchmark scores across Berkeley Function Calling Leaderboard, ARC-AGI-1, LiveCodeBench Pro, SWE-bench Pro, SWE-bench Verified, MMMU-Pro, GPQA Diamond, FrontierMath, and 38 more — each shown with its benchmark, metric, source status, and date, and never combined into a single ranking.

46 results

Benchmark results 46

Compare this model
BenchmarkCategoryScoreMetricStatusDate
Berkeley Function Calling LeaderboardTool use73.24%accuracyOfficialSep 29, 2025Details
ARC-AGI-1Reasoning63.67%accuracyOfficialSep 29, 2025Details
LiveCodeBench ProCoding1412Codeforces EloOfficialSep 29, 2025Details
SWE-bench ProCoding43.60%% resolvedOfficialSep 29, 2025Details
SWE-bench VerifiedCoding71.3%% resolvedOfficialSep 29, 2025Details
MMMU-ProMultimodal68.9%accuracyOfficialSep 29, 2025Details
GPQA DiamondReasoning82.3%accuracyOfficialSep 29, 2025Details
FrontierMathReasoning15.22%accuracyOfficialSep 29, 2025Details
AIME (OTIS Mock)Reasoning77.8%accuracyOfficialSep 29, 2025Details
SimpleQA VerifiedOther23.6%accuracyOfficialSep 29, 2025Details
ARC-AGI-2Reasoning13.61%accuracyOfficialSep 29, 2025Details
Humanity's Last ExamReasoning13.7%accuracyOfficialSep 29, 2025Details
LMArenaChat preference1437source-defined ratingOfficialSep 29, 2025Details
WeirdMLCoding47.7%average accuracyOfficialSep 29, 2025Details
MCP AtlasTool use43.8%pass rateOfficialSep 29, 2025Details
PostTrainBenchAgents9.94%weighted average scoreOfficialSep 29, 2025Details
Epoch Capabilities IndexReasoning147.2IndexOfficialSep 29, 2025Details
MMLU-ProReasoning87.5%accuracyVerifiedSep 29, 2025Details
OSWorldAgents61.4%task success rateUnverifiedSep 29, 2025Details
GAIA: A Benchmark for General AI AssistantsAgents74.55%accuracyUnverifiedSep 29, 2025Details
BrowseCompAgents24.1%accuracyUnverifiedSep 29, 2025Details
GDPvalAgents1317EloOfficialSep 29, 2025Details
LiveCodeBenchCoding59.0%Pass@1UnverifiedSep 29, 2025Details
METR Task-Completion Time HorizonsAgents122 min50% time horizonOfficialSep 29, 2025Details
SciCodeCoding44.7%accuracyUnverifiedSep 29, 2025Details
MultiChallengeReasoning55.32%accuracyVerifiedSep 29, 2025Details
OpenAI-MRCR v2 (Multi-Round Coreference Resolution)Reasoning47.1%accuracy (mean SequenceMatcher similarity)UnverifiedSep 29, 2025Details
Global-MMLUReasoning89.3%accuracyUnverifiedSep 29, 2025Details
WebDev ArenaChat preference1386EloVerifiedSep 29, 2025Details
Search ArenaChat preference1152EloVerifiedSep 29, 2025Details
EQ-Bench Creative Writing v3Chat preference1767EloVerifiedSep 29, 2025Details
Design ArenaChat preference1235EloVerifiedSep 29, 2025Details
MASK (Model Alignment between Statements and Knowledge)Other96.13Honesty scoreVerifiedSep 29, 2025Details
MCP-UniverseTool use35.06%Overall Success RateVerifiedSep 29, 2025Details
ScreenSpot-ProMultimodal36.2%accuracyVerifiedSep 29, 2025Details
FACTS GroundingReasoning61.8%Grounding accuracyVerifiedSep 29, 2025Details
SWE-bench MultilingualCoding67.0%% resolvedOfficialSep 29, 2025Details
EnigmaEvalReasoning6.00%accuracyVerifiedSep 29, 2025Details
ZeroBenchMultimodal0.0% (pass@1)accuracyVerifiedSep 29, 2025Details
FrontierMath Tier 4Reasoning4.2%accuracyOfficialSep 29, 2025Details
Vectara Hallucination LeaderboardOther12.0%Hallucination RateOfficialSep 29, 2025Details
Gray Swan Arena (Agent Red-Teaming / Indirect Prompt Injection)Agents1.0%Attack Success Rate (ASR)VerifiedSep 29, 2025Details
Vibe Code BenchCoding22.62%Overall accuracyVerifiedSep 29, 2025Details
WebArenaAgents58.5%Task success rateVerifiedSep 29, 2025Details
GSO: Software Optimization Benchmark for SWE-AgentsCoding14.71%Opt@1OfficialSep 29, 2025Details
MultiNRCReasoning35.83%accuracyOfficialSep 29, 2025Details