supermarket

Vision-language

Claude Sonnet 4.5

By Anthropic, in the Claude Sonnet family.

Released
29 Sep 2025
Status
Legacy
Context
200K
Max output
64K
Input per 1M tokens
$3.00
Output per 1M tokens
$15.00
Open weights
No
License
proprietary
Parameters
—
Knowledge cutoff
Jan 2025
Reasoning
Yes
Modalities
text, image → text

Category scores

Out of 100, from independent and third-party results only, by method category-2026-09-29.1.

Category scores
CategoryScoreRankCoverage
Knowledge & reasoning25.1694 of 4
Coding & software engineering14.7Not ranked1 of 5
Agentic tool use52.6Not ranked1 of 2
Vision & multimodal understanding9.7Not ranked1 of 3
Factuality & hallucination30.7521 of 1
Human preference54.2193 of 5

Benchmark results

Every result we hold for this model, with who ran it and where it was published.

Knowledge & reasoning

Knowledge & reasoning results
BenchmarkMetricValueProvenanceSettingEvaluatedSource
Chess PuzzlesPuzzles solved12.0%Independent32K8 Dec 2025Epoch AI
Chess PuzzlesPuzzles solved4.0%Independent—6 Aug 2026Epoch AI
DTBench (Decision Theory Benchmark)Accuracy (capability questions)83.2%Third partyunknown—https://conceptualreasoning.ai/dtbench via Epoch AI
Earthborne Rangers (EBR-bench)Objectives reached2.4%Independent—25 Jun 2026Epoch AI
LMCA (Language Model Conceptual Argumentation)Correlation with expert ratings (Pearson r)0.3876Third partyunknown—https://conceptualreasoning.ai/lmca via Epoch AI
Mystery Game PuzzlesPuzzles solved17.0%Independent48K25 Jul 2026Epoch AI
SimpleBenchScore (AVG@5)54.3%Third partyunknown—SimpleBench Leaderboard via Epoch AI
SimpleBenchScore (AVG@5)54.3%Third party12K—SimpleBench Leaderboard via Epoch AI

Scientific reasoning

Scientific reasoning results
BenchmarkMetricValueProvenanceSettingEvaluatedSource
GPQA DiamondAccuracy82.3%Independent59K28 Oct 2025Epoch AI
GPQA DiamondAccuracy81.7%Independent32K21 Oct 2025Epoch AI
GPQA DiamondAccuracy78.8%Independent16K28 Oct 2025Epoch AI
GPQA DiamondAccuracy73.7%Independent—29 Sep 2025Epoch AI

Mathematics

Mathematics results
BenchmarkMetricValueProvenanceSettingEvaluatedSource
FrontierMath Tier 4Accuracy2.4%Independent32K11 Jun 2026Epoch AI
FrontierMath Tier 4 (v1)Accuracy4.2%Independent32K22 Oct 2025Epoch AI
FrontierMath Tier 4 (v1)Accuracy2.1%Independent—29 Sep 2025Epoch AI
FrontierMath Tiers 1-3Accuracy23.9%Independent32K11 Jun 2026Epoch AI
FrontierMath Tiers 1-3 (v1)Accuracy15.2%Independent32K16 Nov 2025Epoch AI
FrontierMath Tiers 1-3 (v1)Accuracy13.5%Independent59K13 Nov 2025Epoch AI
FrontierMath Tiers 1-3 (v1)Accuracy9.3%Independent—16 Nov 2025Epoch AI
MATH Level 5Accuracy97.7%Independent32K21 Oct 2025Epoch AI
OTIS Mock AIME 2024-2025Accuracy77.8%Independent32K21 Oct 2025Epoch AI
OTIS Mock AIME 2024-2025Accuracy77.8%Independent59K28 Oct 2025Epoch AI
OTIS Mock AIME 2024-2025Accuracy71.1%Independent16K28 Oct 2025Epoch AI
OTIS Mock AIME 2024-2025Accuracy35.6%Independent—29 Sep 2025Epoch AI

Coding & software engineering

Coding & software engineering results
BenchmarkMetricValueProvenanceSettingEvaluatedSource
CybenchUnguided challenges solved55.0%Third party——Cybench leaderboard via Epoch AI
CybenchUnguided challenges solved60.0%Self-reportedunknown—Opus 4.5 System Card via Epoch AI
GSOOPT@114.7%Third partyunknown—GSO Leaderboard via Epoch AI
GSOOPT@114.7%Third party——GSO Leaderboard via Epoch AI
METR Time HorizonsAverage task success (Epoch's column)67.4%Third party16K—METR - Measuring AI Ability to Complete Long Tasks via Epoch AI
SWE-bench Verified% resolved71.3%Independent—5 Feb 2026Epoch AI
Terminal-Bench 2.0Accuracy (mean over trials)46.5%Third partyunknown—https://www.tbench.ai/leaderboard/terminal-bench/2.0 via Epoch AI
Terminal-Bench 2.0Accuracy (mean over trials)43.1%Third partyunknown—https://www.tbench.ai/leaderboard/terminal-bench/2.0 via Epoch AI
Terminal-Bench 2.0Accuracy (mean over trials)42.8%Third partyunknown—https://www.tbench.ai/leaderboard/terminal-bench/2.0 via Epoch AI
Terminal-Bench 2.0Accuracy (mean over trials)42.8%Third party——Terminal-Bench v2 Leaderboard via Epoch AI
Terminal-Bench 2.0Accuracy (mean over trials)42.7%Third partyunknown—https://www.tbench.ai/leaderboard/terminal-bench/2.0 via Epoch AI
Terminal-Bench 2.0Accuracy (mean over trials)42.6%Third party——Terminal-Bench v2 Leaderboard via Epoch AI
Terminal-Bench 2.0Accuracy (mean over trials)42.6%Third partyunknown—https://www.tbench.ai/leaderboard/terminal-bench/2.0 via Epoch AI
Terminal-Bench 2.0Accuracy (mean over trials)42.5%Third partyunknown—https://www.tbench.ai/leaderboard/terminal-bench/2.0 via Epoch AI
Terminal-Bench 2.0Accuracy (mean over trials)42.5%Third party——Terminal-Bench v2 Leaderboard via Epoch AI
Terminal-Bench 2.0Accuracy (mean over trials)40.1%Third party——Terminal-Bench v2 Leaderboard via Epoch AI
Terminal-Bench 2.0Accuracy (mean over trials)40.1%Third partyunknown—https://www.tbench.ai/leaderboard/terminal-bench/2.0 via Epoch AI
Terminal-Bench 2.0Accuracy (mean over trials)42.7%Source not statedunknown—Epoch AI via Epoch AI
WeirdML (v2)Average accuracy47.7%Third party16K—https://htihle.github.io/weirdml.html via Epoch AI
WeirdML (v2)Average accuracy46.7%Third party——https://htihle.github.io/weirdml.html via Epoch AI

Agentic tool use

Agentic tool use results
BenchmarkMetricValueProvenanceSettingEvaluatedSource
Deep Research Bench (FutureSearch)Average score52.6%Third party2K—DeepResearchBench Leaderboard via Epoch AI
Deep Research Bench (FutureSearch)Average score47.5%Third partylow—https://evals.futuresearch.ai/#drb via Epoch AI

Computer & browser use

Computer & browser use results
BenchmarkMetricValueProvenanceSettingEvaluatedSource
OSWorld-Verified% tasks completed62.9%Third party——OS World Website via Epoch AI
OSWorld-Verified% tasks completed58.1%Third party——OS World Website via Epoch AI
OSWorld-Verified% tasks completed42.9%Third party——OS World Website via Epoch AI

Professional knowledge work

Professional knowledge work results
BenchmarkMetricValueProvenanceSettingEvaluatedSource
GDPvalWins vs experts42.5%Source not stated——Epoch AI via Epoch AI

Vision & multimodal understanding

Vision & multimodal understanding results
BenchmarkMetricValueProvenanceSettingEvaluatedSource
Visual Physics Comprehension Test (VPCT)Correct39.8%Third party32K—VPCT leaderboard via Epoch AI
Visual Physics Comprehension Test (VPCT)Correct38.0%Third party——VPCT leaderboard via Epoch AI

Factuality & hallucination

Factuality & hallucination results
BenchmarkMetricValueProvenanceSettingEvaluatedSource
SimpleQA VerifiedAccuracy (% correct)30.7%Independent59K27 Aug 2026Epoch AI
SimpleQA VerifiedAccuracy (% correct)23.7%Independent—10 Aug 2026Epoch AI

Human preference

Human preference results
BenchmarkMetricValueProvenanceSettingEvaluatedSource
Code Arena: WebDev (arena.ai)Arena score (Elo scale)1384Independent—25 Sep 2026Arena
Code Arena: WebDev (arena.ai)Votes15728Independent—25 Sep 2026Arena
Document Arena (arena.ai)Arena score (Elo scale)1449Independent—13 Sep 2026Arena
Document Arena (arena.ai)Votes31455Independent—13 Sep 2026Arena
Text Arena (arena.ai)Arena score (Elo scale)1455Independent—25 Sep 2026Arena
Text Arena (arena.ai)Votes82451Independent—25 Sep 2026Arena

Other Claude Sonnet versions

Sources