The Model Index
Up to four models

Claude 3.7 Sonnet vs Claude Sonnet 4

On the index (GPT-4 = 100), Claude 3.7 Sonnet scores 135 and Claude Sonnet 4 133. Side by side on the same third-party tests, with price and context. The link in your address bar shares this exact comparison.

Next to Claude 3.7 Sonnet
Presets
Claude 3.7 Sonnet
135119–151
WeightsClosed
ReleasedFeb 2025
Price in / out–
Context–
Core tests taken4 of 9
Claude Sonnet 4
133117–149
WeightsClosed
ReleasedMay 2025
Price in / out–
Context–
Core tests taken4 of 9

Claude 3.7 Sonnet and Claude Sonnet 4 are too close to call (2 points apart, within the uncertainty), ahead on 2 of 3 shared core tests.

Test by testthird-party results

Claude 3.7 SonnetClaude Sonnet 4
0%25%50%75%100%REASONING & KNOWLEDGEGPQA Diamond (Epoch AI)GPQA Diamond (Epoch AI)Humanity's Last ExamHumanity's Last ExamGPQA Diamond (Vals AI)GPQA Diamond (Vals AI)MMLU ProMMLU ProMMMU ProMMMU ProFiction.LiveBenchFiction.LiveBenchGeoBenchGeoBenchSimpleBenchSimpleBenchVPCTVPCTMATHSAIME-style maths (OTIS mock)AIME-style maths (OTIS mock)MATH (level 5)MATH (level 5)MATH 500MATH 500MGSMMGSMFrontierMath (2025 set)FrontierMath (2025 set)CODINGAider PolyglotAider PolyglotLiveCodeBenchLiveCodeBenchGSO-BenchGSO-BenchAGENTSDeepResearch BenchDeepResearch BenchPROFESSIONAL WORKCorpFin v2CorpFin v2LegalBenchLegalBenchMortgageTaxMortgageTaxTaxEval v2TaxEval v2
Every result as a table
TestClaude 3.7 SonnetClaude Sonnet 4
Reasoning & knowledge
GPQA Diamond (Epoch AI)79.7%79.2%
Humanity's Last Exam8.0%7.8%
GPQA Diamond (Vals AI)75.3%75.0%
MMLU Pro82.7%83.9%
MMMU Pro75.1%74.9%
Fiction.LiveBench83.3%46.9%
GeoBench68.0%37.0%
SimpleBench46.4%45.5%
VPCT39.0%34.0%
Maths
AIME-style maths (OTIS mock)57.8%71.1%
MATH (level 5)91.2%84.4%
MATH 50091.6%93.8%
MGSM93.0%93.0%
FrontierMath (2025 set)4.1%4.1%
Coding
Aider Polyglot64.9%61.3%
LiveCodeBench60.4%62.4%
GSO-Bench3.8%4.9%
Agents
DeepResearch Bench43.6%46.6%
Professional work
CorpFin v260.4%61.2%
LegalBench82.5%83.0%
MortgageTax68.7%62.5%
TaxEval v274.0%72.0%
Task length · METR36 sec37 sec

Tests at least two of these models have taken. Bold rows are the core tests.

This page as a chartDownload PNG