llmboard.aiAI model intelligence
Home

Model Rankings

OverallOpen ModelsAgentCodingReasoningMathKnowledgeInstruction FollowingTextVision
Image GenerationImage Editing
Video GenerationImage to VideoVideo Editing
Text to SpeechSpeech to Text
Embeddings

Efficiency

Chat Token PricingImage PricingVideo PricingAudio Pricing
Chat Speed & LatencyProvider Reliability

Benchmarks

GPQAMMLU-ProAIME 2025SWE-Bench VerifiedMMLUHumanity's Last ExamLiveCodeBenchMATHHumanEvalMMMU-Pro
All Benchmarks

Tools

Model Directory

Scoring & Data

Scoring & Data
1224 models729 benchmarks

Leaderboard Center

Overall RankingCodingCore BenchmarksPrice & ValueRuntime Performance

Modalities

All ModelsImage GenerationImage EditingVideo GenerationImage-to-VideoVideo EditingText-to-SpeechSpeech-to-TextEmbeddings

Data & Methods

Scoring MethodAll Benchmarks
llmboard.aiCopyright 2026 llmboard.ai

chat benchmark

Arena Hard Leaderboard

Arena Hard is an automatic evaluation benchmark for instruction-tuned LLMs based on 500 challenging real-world prompts curated by BenchBuilder.

Updated Sep 5, 2026

Models26
Model coverage26
MetricScore
EvidenceB

On this page

  • Ranking
  • Highlights
  • Distribution
  • Top models
  • About
  • FAQ

Arena Hard Ranking

Higher score ranks better on this benchmark.

26 rows
Columns

Show columns

Sort by
Rank
Model
Score
Percentile
Participants
Evidence
Evaluated
Rank01ModelACQwen3 235B A22BAlibaba Cloud / Qwen TeamScore95.60%Percentile100.00%Participants26EvidenceCEvaluatedSep 8, 2026
Rank02ModelACQwen3 32BAlibaba Cloud / Qwen TeamScore93.80%Percentile96.00%Participants26EvidenceCEvaluatedSep 8, 2026
Rank03ModelACQwen3 30B A3BAlibaba Cloud / Qwen TeamScore91.00%Percentile92.00%Participants26EvidenceCEvaluatedSep 8, 2026
Rank04ModelNVLlama-3.3 Nemotron Super 49B v1NVIDIAScore88.30%Percentile88.00%Participants26EvidenceCEvaluatedSep 8, 2026
Rank05ModelMAMistral Small 3 24B InstructMistral AIScore87.60%Percentile84.00%Participants26EvidenceCEvaluatedSep 8, 2026
Rank06ModelACQwen2.5 72B InstructAlibaba Cloud / Qwen TeamScore81.20%Percentile80.00%Participants26EvidenceCEvaluatedSep 8, 2026
Rank07ModelMIPhi 4 Reasoning PlusMicrosoftScore79.00%Percentile76.00%Participants26EvidenceCEvaluatedSep 8, 2026
Rank08ModelDEDeepSeek-V2.5DeepSeekScore76.20%Percentile72.00%Participants26EvidenceCEvaluatedSep 8, 2026
Rank09ModelMIPhi 4MicrosoftScore75.40%Percentile68.00%Participants26EvidenceCEvaluatedSep 8, 2026
Rank10ModelMIPhi 4 ReasoningMicrosoftScore73.30%Percentile64.00%Participants26EvidenceCEvaluatedSep 8, 2026
Rank11ModelMAMinistral 8B InstructMistral AIScore70.90%Percentile60.00%Participants26EvidenceCEvaluatedSep 8, 2026
Rank12ModelALJamba 1.5 LargeAI21 LabsScore65.40%Percentile56.00%Participants26EvidenceCEvaluatedSep 8, 2026
Rank13ModelMAMistral Small 4Mistral AIScore58.30%Percentile52.00%Participants26EvidenceCEvaluatedSep 8, 2026
Rank14ModelIBGranite 3.3 8B BaseIBMScore57.56%Percentile48.00%Participants26EvidenceCEvaluatedSep 8, 2026
Rank15ModelIBGranite 3.3 8B InstructIBMScore57.56%Percentile44.00%Participants26EvidenceCEvaluatedSep 8, 2026
Rank16ModelMAMiniStral 3 (14B Instruct 2512)Mistral AIScore55.10%Percentile40.00%Participants26EvidenceCEvaluatedSep 8, 2026
Rank17ModelMAMistral Large 3Mistral AIScore55.10%Percentile36.00%Participants26EvidenceCEvaluatedSep 8, 2026
Rank18ModelACQwen2.5 7B InstructAlibaba Cloud / Qwen TeamScore52.00%Percentile32.00%Participants26EvidenceCEvaluatedSep 8, 2026
Rank19ModelMAMinistral 3 (8B Instruct 2512)Mistral AIScore50.90%Percentile28.00%Participants26EvidenceCEvaluatedSep 8, 2026
Rank20ModelALJamba 1.5 MiniAI21 LabsScore46.10%Percentile24.00%Participants26EvidenceCEvaluatedSep 8, 2026
Rank21ModelMAMistral Small 3.2 24B InstructMistral AIScore43.10%Percentile20.00%Participants26EvidenceCEvaluatedSep 8, 2026
Rank22ModelMIPhi-3.5-MoE-instructMicrosoftScore37.90%Percentile16.00%Participants26EvidenceCEvaluatedSep 8, 2026
Rank23ModelMIPhi-3.5-mini-instructMicrosoftScore37.00%Percentile12.00%Participants26EvidenceCEvaluatedSep 8, 2026
Rank24ModelMIPhi 4 MiniMicrosoftScore32.80%Percentile8.00%Participants26EvidenceCEvaluatedSep 8, 2026
Rank25ModelMAMinistral 3 (3B Instruct 2512)Mistral AIScore30.50%Percentile4.00%Participants26EvidenceCEvaluatedSep 8, 2026
Rank26ModelIBIBM Granite 4.0 Tiny PreviewIBMScore26.70%Percentile0.00%Participants26EvidenceCEvaluatedSep 8, 2026

Arena Hard Highlights

The leading models and scores on this benchmark.

Arena Hard Score Distribution

A closer view of the leading scores on this benchmark.

Arena Hard

The Top AI Models for Arena Hard

The first five results on this benchmark, with official price and output speed added where the model identity can be matched.

What is Arena Hard?

What Arena Hard measures and how its scores work.

Arena-Hard-Auto is a benchmark comprising open-ended software engineering problems, mathematical questions, and creative writing tasks, evaluated with GPT-4.1 and Gemini-2.5 as automatic judges.

It measures model performance using LLM-as-a-Judge evaluations that approximate human preference, reported as a Score with unit ratio.

Scores are shown in ratio. This benchmark is not independently verified and has an evidence level of B.

Family
Arena Hard
Modality
text
Primary category
chat
Score direction
higher
LLMBoard eligible
No
Evaluation key
overall

LLM Stats. Benchmark scores retain their original unit. Overall score eligibility is shown separately.

FAQ

Common questions about Arena Hard.

Which model scores highest on Arena Hard?

Qwen3 235B A22B is currently ranked first with 95.60%.

What are the top three models on Arena Hard?

The current leaders are Qwen3 235B A22B (95.60%), Qwen3 32B (93.80%), and Qwen3 30B A3B (91.00%).

Which Arena Hard model has the lowest official input price?

Phi 4 Mini has the lowest matched official input price at $0.08 input / $0.30 output per 1M tokens.

Which models are fastest among Arena Hard results?

The fastest matched records are Mistral Small 4 (163.71 tok/s via Mistral AI), Mistral Small 3 24B Instruct (134.00 tok/s via Mistral AI), and DeepSeek-V2.5 (100.00 tok/s via DeepSeek).

Does the highest score result prove overall model quality?

No. This benchmark measures one defined capability or task. The overall LLMBoard score uses a separate aggregation across eligible benchmark evidence.

What does Arena Hard measure?

It measures model performance using LLM-as-a-Judge evaluations that approximate human preference, reported as a Score with unit ratio.

Is a higher score better?

Yes. Higher values rank better for this benchmark.

How many models are compared?

26 model results are currently shown.

Does this benchmark affect the overall score?

No. This benchmark is shown for reference but does not contribute to the overall score.

Rank #1Qwen3 235B A22B95.60%
Rank #2Qwen3 32B93.80%
Rank #3Qwen3 30B A3B91.00%
Rank #4Llama-3.3 Nemotron Super 49B v188.30%

Ranking basisThis arena hard AI model leaderboard uses descending score in the benchmark original unit. The leaderboard ranking keeps matched price and speed data separate from benchmark evidence.

  1. 01
    AC
    Qwen3 235B A22BAlibaba Cloud / Qwen Team
    Score
    95.60%
    Price
    $0.70 input / $2.8 output per 1M tokens
    Speed
    Up to 21.74 tok/s via DeepInfra

    Strengths

    • Ranks #1 of 26 compared models
    • 100th percentile on this benchmark
    • C evidence result

    Considerations

    • This result measures Arena Hard, not total model capability
  2. 02
    AC
    Qwen3 32BAlibaba Cloud / Qwen Team
    Score
    93.80%
    Price
    $0.70 input / $2.8 output per 1M tokens
    Speed
    Up to 26.95 tok/s via DeepInfra

    Strengths

    • Ranks #2 of 26 compared models
    • 96th percentile on this benchmark
    • C evidence result

    Considerations

    • This result measures Arena Hard, not total model capability
  3. 03
    AC
    Alibaba Cloud / Qwen Team
    Score
    91.00%
    Speed
    Up to 82.57 tok/s via DeepInfra

    Strengths

    • Ranks #3 of 26 compared models
    • 92th percentile on this benchmark
    • C evidence result

    Considerations

    • This result measures Arena Hard, not total model capability
  4. 04
    NV
    NVIDIA
    Score
    88.30%

    Strengths

    • Ranks #4 of 26 compared models
    • 88th percentile on this benchmark
    • C evidence result

    Considerations

    • This result measures Arena Hard, not total model capability
  5. 05
    MA
    Mistral AI
    Score
    87.60%
    Speed
    Up to 134.00 tok/s via Mistral AI

    Strengths

    • Ranks #5 of 26 compared models
    • 84th percentile on this benchmark
    • C evidence result

    Considerations

    • This result measures Arena Hard, not total model capability

Selection summary

Best AI Models for Arena Hard

Qwen3 235B A22B currently leads Arena Hard with 95.60%. It is the top model on this specific benchmark, while the best LLM for the broader task should also be checked against other benchmarks, price, and runtime.

Use this leaderboard with the supporting benchmark results and coverage details above. A leaderboard position summarizes the selected ranking signal; it does not replace workload-specific testing.

Qwen3 30B A3B
Llama-3.3 Nemotron Super 49B v1
Mistral Small 3 24B Instruct
Benchmark rank #1Qwen3 235B A22B95.60% · $0.70 input / $2.8 output per 1M tokens
Benchmark rank #2Qwen3 32B93.80% · $0.70 input / $2.8 output per 1M tokens
Benchmark rank #3Qwen3 30B A3B91.00% · Up to 82.57 tok/s via DeepInfra