AIEvals

aievals.app

Who leads what

55 independent leaderboards, read 17 October 2026. Each stays tied to its publisher and date instead of being averaged into one score.

Individual benchmark leaderboards
55of 128 benchmarks on this site
Labs holding a first place
7ties counted for both labs
Models compared
46every model with a page on this site
Publisher-reported only
71benchmarks with no independent suite

First place, benchmark by benchmark

The latest top two from each independent leaderboard. A dash means no runner-up is published yet.

Independent benchmark leaders, next listed models, scores, field sizes and update dates.
BenchmarkLeaderScoreNext listed modelScoreModels runUpdated
APEX-Agents-AAArtificial Analysis · Knowledge workGemini 3.5 Flash47.05%Qwen3.8 Max (0902)Alibaba42.40%32spread 15.04%30 Sept 2026
τ³-BankingArtificial Analysis · Tool useQwen3.8 MaxAlibaba51.34%Muse Spark 1.3Meta50.52%154spread 5.98%30 Sept 2026
ITBench-AAArtificial Analysis · IT operationsGPT-5.6 SolOpenAI56.21%Gemini 3.8 FlashGoogle52.54%42spread 9.55%30 Sept 2026
IFBenchArtificial Analysis · Instruction followingGrok 4.20 030982.93%MiniMax-M3MiniMax82.86%331spread 4.15%30 Sept 2026
AIME 2025Artificial Analysis · MathematicsGPT-5.299.00%GPT-5 Codex98.67%208spread 4.67%30 Sept 2026
AA-AnalystAgentArtificial Analysis · Data analysisGemini 3.7 FlashGoogle60.00%Claude Fable 5.1Anthropic57.50%33spread 15.00%30 Sept 2026
MLCR-AAArtificial Analysis · Medical reasoningClaude Sonnet 5.5Anthropic75.00%Claude Fable 5.1Anthropic71.11%70spread 31.67%30 Sept 2026
ARC-AGI-3ARC Prize · Standard · Interactive reasoningGPT-6 Astra (Max)OpenAI62.71%Claude Opus 5 (High)Anthropic30.16%17spread 62.28%30 Sept 2026
SWE-Bench Pro V2 FullScale AI / SEAL · Repository codingOpus 5 (Claude Code) xhighAnthropic99.40%Fable 5.1 (Claude Code) highAnthropic99.10%10spread 9.52%22 Sept 2026
SWE-Bench Pro V2 HARDScale AI / SEAL · Repository codingOpus 5 (Claude Code) xhighAnthropic98.00%Fable 5.1 (Claude Code) highAnthropic92.20%11spread 41.10%22 Sept 2026
Vals RSI IndexVals AI · Autonomous researchClaude Opus 5.5Anthropic37.31%Claude Fable 5.1Anthropic36.09%23spread 13.43%29 Sept 2026
Vibe Code Bench 1–100Vals AI · Application developmentClaude Opus 5.5Anthropic30.36%Claude Opus 5Anthropic28.53%20spread 12.12%29 Sept 2026
CUA-benchVals AI · Computer useGPT-6 AstraOpenAI19.17%Claude Opus 5.5Anthropic14.00%8spread 15.00%29 Sept 2026
τ²-Bench TelecomArtificial Analysis · Tool useJT-35B-Flash99.12%GLM-5.2Tied first99.12%323spread 1.46%30 Sept 2026
Harvey LAB-AAArtificial Analysis · Legal reasoningMuse Spark 1.3Meta95.48%Kimi K3Moonshot AI94.64%42spread 4.40%8 Oct 2026
EnterpriseOps-Gym-AAArtificial Analysis · Enterprise workflowsClaude Fable 5Anthropic51.12%Gemini 3.7 FlashGoogle50.40%47spread 4.03%30 Sept 2026
AutomationBench-AAArtificial Analysis · Workflow automationGemini 4 ArgonGoogle77.51%Claude Sonnet 5.5Anthropic71.33%140spread 15.31%8 Oct 2026
SciCodeArtificial Analysis · Scientific codingClaude Opus 5.5Anthropic66.90%Claude Fable 5.1Anthropic63.08%144spread 8.10%8 Oct 2026
GDP.pdfArtificial Analysis · Document reasoningGPT-6.1 SolOpenAI31.00%GPT-6 AstraOpenAITied first31.00%130spread 7.00%8 Oct 2026
CritPtArtificial Analysis · Scientific reasoningGPT-5.6 SolOpenAI32.29%GPT-6.1 SolOpenAI31.71%386spread 2.58%8 Oct 2026
AA-OmniscienceArtificial Analysis · Factual knowledgeClaude Opus 5.5Anthropic46.42Claude Fable 5.1Anthropic43.45383spread 14.548 Oct 2026
AA-LCR v1.1Artificial Analysis · Long-context reasoningKimi K3Moonshot AI88.67%Step 5 PreviewStepFun88.33%386spread 5.00%8 Oct 2026
Terminal-Bench 4.0Artificial Analysis · Agentic codingClaude Sonnet 5.5Anthropic63.64%Claude Opus 5.5Anthropic59.60%137spread 21.72%8 Oct 2026
Terminal-Bench 2.1Vals AI · Agentic codingClaude Opus 5.5Anthropic87.64%GPT-6 AstraOpenAI87.27%75spread 7.12%27 Sept 2026
SWE-bench (Verified)Vals AI · Software engineeringClaude Opus 5Anthropic97.00%DeepSeek V4 ProDeepSeek96.40%88spread 5.00%1 Sept 2026
LiveCodeBenchVals AI · Code generationClaude Fable 5.1Anthropic90.52%Claude Fable 5Anthropic89.78%143spread 2.67%1 Sept 2026
GPQA DiamondVals AI · Science reasoningGemini 3.1 ProGoogle95.45%GPT-5.6 SolOpenAI95.20%138spread 2.27%1 Sept 2026
MMLU ProVals AI · KnowledgeClaude Fable 5.1Anthropic92.38%Claude Opus 5Anthropic91.59%138spread 2.86%1 Sept 2026
MMMU ProVals AI · Multimodal reasoningClaude Fable 5.1Anthropic90.64%Claude Opus 5Anthropic89.88%93spread 2.43%1 Sept 2026
Vibe Code Bench v1.1Vals AI · Product codingClaude Sonnet 5.5Anthropic92.39%Gemini 4 ArgonGoogle91.91%106spread 6.22%29 Sept 2026
Terminal-Bench Science 0.1Artificial Analysis · Agentic scienceGPT-6 AstraOpenAI63.33%Claude Opus 5.5Anthropic59.05%27spread 49.52%8 Oct 2026
ProgramBenchVals AI · Agentic codingClaude Opus 5.5Anthropic18.50%Claude Fable 5.1Anthropic7.00%54spread 17.00%29 Sept 2026
ProofBench v1.1Vals AI · MathematicsClaude Sonnet 5.5Anthropic100.00%Claude Opus 5.5AnthropicTied first100.00%46spread 13.00%29 Sept 2026
LegalBenchVals AI · Professional knowledgeClaude Fable 5Anthropic88.56%Claude Fable 5.1Anthropic88.51%149spread 1.70%29 Sept 2026
Harvey's Legal Agent BenchmarkVals AI · Professional knowledgeMuse Spark 1.2Meta25.42%Muse Spark 1.3 MaxMeta23.75%73spread 14.17%30 Sept 2026
SkillsBenchVals AI · Agentic toolingDeepSeek V4.1 FlashDeepSeek69.80%Grok 4.5xAI66.03%35spread 10.61%27 Sept 2026
MysteryMechanismVals AI · Scientific reasoningGPT-6 AstraOpenAI53.15%Claude Opus 5.5Anthropic49.55%21spread 19.82%29 Sept 2026
Vals Multimodal IndexVals AI · CompositeClaude Fable 5Anthropic74.15%Claude Opus 5Anthropic73.90%33spread 6.79%11 Aug 2026
BioMysteryBenchVals AI · Agentic scienceClaude Sonnet 5.5Anthropic81.11%GPT-6.1 SolOpenAI79.63%21spread 10.00%29 Sept 2026
Code MigrationVals AI · Agentic codingClaude Sonnet 5.5Anthropic69.83%Gemini 4 ArgonGoogle68.17%71spread 16.91%29 Sept 2026
CyberBench v1.1Vals AI · SecurityGPT-6 SolOpenAI77.98%Gemini 4 ArgonGoogle77.86%43spread 5.00%29 Sept 2026
EMBVals AI · Financial modellingClaude Fable 5.1Anthropic76.67%Claude Opus 5.5Anthropic75.94%69spread 5.14%29 Sept 2026
Finance Agent (v2)Vals AI · Financial analysisGemini 4 ArgonGoogle65.40%Gemini 3.8 FlashGoogle61.44%73spread 7.30%29 Sept 2026
IOIVals AI · Competitive programmingGemini 4 ArgonGoogle100.00%GPT-6 AstraOpenAITied first100.00%38spread 17.39%29 Sept 2026
Legal Research BenchVals AI · Professional knowledgeMuse Spark 1.3 MaxMeta55.29%Claude Opus 5AnthropicTied first55.29%72spread 7.21%29 Sept 2026
MedCodeVals AI · Healthcare administrationClaude Opus 5Anthropic63.57%Gemini 3.1 ProGoogle59.06%104spread 10.35%29 Sept 2026
MedScribeVals AI · Healthcare administrationClaude Opus 5.5Anthropic91.43%Claude Fable 5.1Anthropic91.29%106spread 2.91%29 Sept 2026
Public Benefits Bench v1.1Vals AI · Public servicesClaude Opus 5Anthropic76.93%Claude Fable 5.1Anthropic74.90%45spread 8.66%29 Sept 2026
SAGEVals AI · EducationClaude Opus 4.7Anthropic56.10%Gemma 4 31b It55.03%90spread 4.25%29 Sept 2026
SRE BenchVals AI · SecurityGPT-6 AstraOpenAI56.87%GPT-6.1 SolOpenAI50.76%16spread 52.67%29 Sept 2026
Tax Agent BenchVals AI · Professional knowledgeClaude Fable 5.1Anthropic77.64%Gemini 4 ArgonGoogle76.23%64spread 7.82%29 Sept 2026
DeepSWE v1.1Datacurve · Software engineeringGPT-6 AstraOpenAI74.12%Gemini 3.8 FlashGoogle73.83%28spread 6.93%22 Sept 2026
GDPval-AA v2.1Artificial Analysis · Knowledge workClaude Opus 5.5Anthropic1,846.17Claude Sonnet 5.5Anthropic1,844.21188spread 205.398 Oct 2026
AA-Briefcase v1.1Artificial Analysis · Knowledge workClaude Opus 5.5Anthropic1,821.85Claude Sonnet 5.5Anthropic1,810.9142spread 252.998 Oct 2026
Humanity's Last ExamArtificial Analysis · Multidisciplinary reasoningClaude Opus 5.5Anthropic61.35%Claude Fable 5.1Anthropic59.13%466spread 12.00%8 Oct 2026

APEX-Agents-AA

Artificial Analysis · Knowledge work

Winner
Gemini 3.5 Flash47.05%
Next listed
Qwen3.8 Max (0902)Alibaba42.40%
32 modelsSpread 15.04%Updated 30 Sept 2026

IFBench

Artificial Analysis · Instruction following

Winner
Grok 4.20 030982.93%
Next listed
MiniMax-M3MiniMax82.86%
331 modelsSpread 4.15%Updated 30 Sept 2026

AIME 2025

Artificial Analysis · Mathematics

Winner
GPT-5.299.00%
Next listed
GPT-5 Codex98.67%
208 modelsSpread 4.67%Updated 30 Sept 2026

τ²-Bench Telecom

Artificial Analysis · Tool use

Tied first
JT-35B-Flash99.12%
Tied first
GLM-5.299.12%
323 modelsSpread 1.46%Updated 30 Sept 2026

GDP.pdf

Artificial Analysis · Document reasoning

Tied first
GPT-6.1 SolOpenAI31.00%
Tied first
GPT-6 AstraOpenAI31.00%
130 modelsSpread 7.00%Updated 8 Oct 2026

CritPt

Artificial Analysis · Scientific reasoning

Winner
GPT-5.6 SolOpenAI32.29%
Next listed
GPT-6.1 SolOpenAI31.71%
386 modelsSpread 2.58%Updated 8 Oct 2026

AA-LCR v1.1

Artificial Analysis · Long-context reasoning

Winner
Kimi K3Moonshot AI88.67%
Next listed
Step 5 PreviewStepFun88.33%
386 modelsSpread 5.00%Updated 8 Oct 2026

IOI

Vals AI · Competitive programming

Tied first
Gemini 4 ArgonGoogle100.00%
Tied first
GPT-6 AstraOpenAI100.00%
38 modelsSpread 17.39%Updated 29 Sept 2026

SAGE

Vals AI · Education

Winner
Claude Opus 4.7Anthropic56.10%
Next listed
Gemma 4 31b It55.03%
90 modelsSpread 4.25%Updated 29 Sept 2026
Scope: Vals AI and Artificial Analysis. Publisher comparison tables are not counted.

By lab

Independent runs only. Each first place counts once; ties count for both labs.

Labs ordered by first places, their top-three finishes, and the benchmarks they lead.
LabFirst placesTop-three finishesLeading benchmarks
Anthropic3140
28 more benchmarks
OpenAI1120
8 more benchmarks
Google621
3 more benchmarks
Meta35
Moonshot AI15
Alibaba13
DeepSeek12
xAI02No benchmark wins
Z.ai02No benchmark wins
MiniMax01No benchmark wins
StepFun01No benchmark wins
Xiaomi01No benchmark wins

Anthropic

First places
31
Top-three
40

Leading benchmarks

28 more benchmarks

Moonshot AI

First places
1
Top-three
5

Leading benchmarks

Alibaba

First places
1
Top-three
3

Leading benchmarks

DeepSeek

First places
1
Top-three
2

Leading benchmarks

xAI

First places
0
Top-three
2

Leading benchmarks

No benchmark wins

Z.ai

First places
0
Top-three
2

Leading benchmarks

No benchmark wins

MiniMax

First places
0
Top-three
1

Leading benchmarks

No benchmark wins

StepFun

First places
0
Top-three
1

Leading benchmarks

No benchmark wins

Xiaomi

First places
0
Top-three
1

Leading benchmarks

No benchmark wins

The two composites, side by side

Read the indices last, not first: each is one publisher’s weighting over its own suite, and the two share no task and no harness.

Artificial Analysis Intelligence Index

Artificial Analysis · 492 models · updated 8 October 2026

#ModelIndex
1Claude Opus 5.558
2Claude Sonnet 5.556
3Claude Fable 5.153
3Gemini 4 Argon53
3GPT-6 Astra53
6GPT-6.1 Sol52
7Claude Opus 551
8Claude Fable 550

Top eight shown. Search to find another model. An index over a fixed suite of reasoning, coding, mathematics and agentic evaluations, reported on its own scale rather than as a percentage.

Artificial Analysis ranks 492 models; this site lists the 42 it read, so these ranks are among those 42 only.

artificialanalysis.ai/leaderboards/models