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 DirectoryCompare Models

Scoring & Data

Scoring & Data
393 models668 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 BenchmarksReasoningMath

Vendors

All VendorsOpenAIAnthropicGoogle
llmboard.aiCopyright 2026 llmboard.ai

reasoning benchmark

Terminal-Bench Leaderboard

Terminal-Bench is a benchmark for testing AI agents in real terminal environments. It evaluates how well agents can handle real-world, end-to-end tasks autonomously, including compiling code, training models, setting up servers, system administration, security tasks, data science workflows, and cybersecurity vulnerabilities. The benchmark consists of a dataset of ~100 hand-crafted, human-verified tasks and an execution harness that connects language models to a terminal sandbox.

Updated Aug 17, 2026

Models25
Model coverage25
MetricScore
EvidenceB

On this page

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

Terminal-Bench Ranking

Higher score ranks better on this benchmark.

25 rows
Columns

Show columns

Sort by
Rank
Model
Score
Percentile
Participants
Evidence
Evaluated
Rank01ModelANClaude Sonnet 4.5AnthropicScore50.0%Percentile100.0%Participants25EvidenceCEvaluatedAug 17, 2026
Rank02ModelMIMiniMax M2.1MiniMaxScore47.9%Percentile95.8%Participants25EvidenceCEvaluatedAug 17, 2026
Rank03ModelMAKimi K2-Thinking-0905Moonshot AIScore47.1%Percentile91.7%Participants25EvidenceCEvaluatedAug 17, 2026
Rank04ModelMIMiniMax M2MiniMaxScore46.3%Percentile87.5%Participants25EvidenceCEvaluatedAug 17, 2026
Rank05ModelANClaude Opus 4.1AnthropicScore43.3%Percentile83.3%Participants25EvidenceCEvaluatedAug 17, 2026
Rank06ModelAMNova 2 ProAmazonScore41.3%Percentile79.2%Participants25EvidenceCEvaluatedAug 17, 2026
Rank07ModelANClaude Haiku 4.5AnthropicScore41.0%Percentile75.0%Participants25EvidenceCEvaluatedAug 17, 2026
Rank08ModelZAGLM-4.6Zhipu AIScore40.5%Percentile70.8%Participants25EvidenceCEvaluatedAug 17, 2026
Rank09ModelMELongCat-Flash-ChatMeituanScore39.5%Percentile66.7%Participants25EvidenceCEvaluatedAug 17, 2026
Rank10ModelANClaude Opus 4AnthropicScore39.2%Percentile62.5%Participants25EvidenceCEvaluatedAug 17, 2026
Rank11ModelDEDeepSeek-V3.2-ExpDeepSeekScore37.7%Percentile58.3%Participants25EvidenceCEvaluatedAug 17, 2026
Rank12ModelZAGLM-4.5Zhipu AIScore37.5%Percentile54.2%Participants25EvidenceCEvaluatedAug 17, 2026
Rank13ModelANClaude Sonnet 4AnthropicScore35.5%Percentile50.0%Participants25EvidenceCEvaluatedAug 17, 2026
Rank14ModelANClaude 3.7 SonnetAnthropicScore35.2%Percentile45.8%Participants25EvidenceCEvaluatedAug 17, 2026
Rank15ModelMELongCat-Flash-LiteMeituanScore33.8%Percentile41.7%Participants25EvidenceCEvaluatedAug 17, 2026
Rank16ModelZAGLM-4.7Zhipu AIScore33.3%Percentile37.5%Participants25EvidenceCEvaluatedAug 17, 2026
Rank17ModelAMNova 2 LiteAmazonScore32.5%Percentile33.3%Participants25EvidenceCEvaluatedAug 17, 2026
Rank18ModelDEDeepSeek-V3.1DeepSeekScore31.3%Percentile29.2%Participants25EvidenceCEvaluatedAug 17, 2026
Rank19ModelXIMiMo-V2-FlashXiaomiScore30.5%Percentile25.0%Participants25EvidenceCEvaluatedAug 17, 2026
Rank20ModelZAGLM-4.5-AirZhipu AIScore30.0%Percentile20.8%Participants25EvidenceCEvaluatedAug 17, 2026
Rank21ModelMAKimi K2 InstructMoonshot AIScore30.0%Percentile16.7%Participants25EvidenceCEvaluatedAug 17, 2026
Rank22ModelNVNemotron 3 Super (120B A12B)NVIDIAScore25.8%Percentile12.5%Participants25EvidenceCEvaluatedAug 17, 2026
Rank23ModelMAKimi K2-Instruct-0905Moonshot AIScore25.0%Percentile8.3%Participants25EvidenceCEvaluatedAug 17, 2026
Rank24ModelNVNemotron 3 Nano (30B A3B)NVIDIAScore8.5%Percentile4.2%Participants25EvidenceCEvaluatedAug 17, 2026
Rank25ModelDEDeepSeek-R1-0528DeepSeekScore5.7%Percentile0.0%Participants25EvidenceCEvaluatedAug 17, 2026

Terminal-Bench Highlights

The leading models and scores on this benchmark.

Rank #1Claude Sonnet 4.550.0%Rank #2MiniMax M2.147.9%Rank #3Kimi K2-Thinking-090547.1%Rank #4MiniMax M246.3%

Terminal-Bench Score Distribution

A closer view of the leading scores on this benchmark.

Terminal-Bench

The Top AI Models for Terminal-Bench

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

Ranking basisThis terminal-bench AI model leaderboard uses descending score in the benchmark's original unit. The leaderboard ranking keeps matched price and speed data separate from benchmark evidence.

  1. 01
    AN
    Claude Sonnet 4.5Anthropic
    Score
    50.0%
    Price
    $3.0 input / $15 output per 1M tokens
    Speed
    Up to 42 tok/s via Anthropic

    Strengths

    • Ranks #1 of 25 compared models
    • 100th percentile on this benchmark
    • C evidence result

    Considerations

    • This result measures Terminal-Bench, not total model capability
  2. 02
    MI
    MiniMax M2.1MiniMax
    Score
    47.9%
    Price
    $0.30 input / $1.2 output per 1M tokens
    Speed
    Up to 100 tok/s via MiniMax

    Strengths

    • Ranks #2 of 25 compared models
    • 96th percentile on this benchmark
    • C evidence result

    Considerations

    • This result measures Terminal-Bench, not total model capability
  3. 03
    MA
    Kimi K2-Thinking-0905Moonshot AI
    Score
    47.1%
    Price
    $0.60 input / $2.5 output per 1M tokens

    Strengths

    • Ranks #3 of 25 compared models
    • 92th percentile on this benchmark
    • C evidence result

    Considerations

    • This result measures Terminal-Bench, not total model capability
  4. 04
    MI
    MiniMax M2MiniMax
    Score
    46.3%
    Price
    $0.30 input / $1.2 output per 1M tokens
    Speed
    Up to 103 tok/s via MiniMax

    Strengths

    • Ranks #4 of 25 compared models
    • 88th percentile on this benchmark
    • C evidence result

    Considerations

    • This result measures Terminal-Bench, not total model capability
  5. 05
    AN
    Claude Opus 4.1Anthropic
    Score
    43.3%
    Speed
    Up to 120 tok/s via Bedrock

    Strengths

    • Ranks #5 of 25 compared models
    • 83th percentile on this benchmark
    • C evidence result

    Considerations

    • This result measures Terminal-Bench, not total model capability

Selection summary

Best AI Models for Terminal-Bench

Claude Sonnet 4.5 currently leads Terminal-Bench with 50.0%. 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.

Benchmark rank #1Claude Sonnet 4.550.0% · $3.0 input / $15 output per 1M tokensBenchmark rank #2MiniMax M2.147.9% · $0.30 input / $1.2 output per 1M tokensBenchmark rank #3Kimi K2-Thinking-090547.1% · $0.60 input / $2.5 output per 1M tokens

What is Terminal-Bench?

What Terminal-Bench measures and how its scores work.

Terminal-Bench is a benchmark for testing AI agents in real terminal environments. It evaluates how well agents can handle real-world, end-to-end tasks autonomously, including compiling code, training models, setting up servers, system administration, security tasks, data science workflows, and cybersecurity vulnerabilities. The benchmark consists of a dataset of ~100 hand-crafted, human-verified tasks and an execution harness that connects language models to a terminal sandbox.

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

Family
Terminal-Bench
Modality
text
Primary category
reasoning
Score direction
higher
LLMBoard eligible
Yes
Evaluation key
terminal-bench|llm-stats-current

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

FAQ

Common questions about Terminal-Bench.

Which model scores highest on Terminal-Bench?

Claude Sonnet 4.5 is currently ranked first with 50.0%.

What does Terminal-Bench measure?

Terminal-Bench is a benchmark for testing AI agents in real terminal environments. It evaluates how well agents can handle real-world, end-to-end tasks autonomously, including compiling code, training models, setting up servers, system administration, security tasks, data science workflows, and cybersecurity vulnerabilities. The benchmark consists of a dataset of ~100 hand-crafted, human-verified tasks and an execution harness that connects language models to a terminal sandbox.

Is a higher score better?

Yes. Higher values rank better for this benchmark.

How many models are compared?

25 model results are currently shown.

Does this benchmark affect the overall score?

Yes. This benchmark can contribute to the current LLMBoard capability score.