reasoning benchmark
Terminal-Bench Leaderboard
Terminal-Bench evaluates AI agents on autonomous, end-to-end tasks in real terminal environments.
Updated Sep 5, 2026
Models25
Model coverage25
MetricScore
EvidenceB
Terminal-Bench Ranking
Higher score ranks better on this benchmark.
Sort by
Rank | Model | Score | Percentile | Participants | Evidence | Evaluated |
|---|
| Rank01 | ModelAN | Score50.00% | Percentile100.00% | Participants25 | EvidenceC | Evaluated |
| Rank02 | ModelMI | Score47.90% | Percentile95.83% | Participants25 | EvidenceC | Evaluated |
| Rank03 | ModelMA | Score47.10% | Percentile91.67% | Participants25 | EvidenceC | Evaluated |
| Rank04 | ModelMI | Score46.30% | Percentile87.50% | Participants25 | EvidenceC | Evaluated |
| Rank05 | ModelAN | Score43.30% | Percentile83.33% | Participants25 | EvidenceC | Evaluated |
| Rank06 | ModelAM Nova 2 ProAmazon | Score41.30% | Percentile79.17% | Participants25 | EvidenceC | Evaluated |
| Rank07 | ModelAN | Score41.00% | Percentile75.00% | Participants25 | EvidenceC | Evaluated |
| Rank08 | ModelZA | Score40.50% | Percentile70.83% | Participants25 | EvidenceC | Evaluated |
| Rank09 | ModelME | Score39.51% | Percentile66.67% | Participants25 | EvidenceC | Evaluated |
| Rank10 | ModelAN | Score39.20% | Percentile62.50% | Participants25 | EvidenceC | Evaluated |
| Rank11 | ModelDE | Score37.70% | Percentile58.33% | Participants25 | EvidenceC | Evaluated |
| Rank12 | ModelZA | Score37.50% | Percentile54.17% | Participants25 | EvidenceC | Evaluated |
| Rank13 | ModelAN | Score35.50% | Percentile50.00% | Participants25 | EvidenceC | Evaluated |
| Rank14 | ModelAN | Score35.20% | Percentile45.83% | Participants25 | EvidenceC | Evaluated |
| Rank15 | ModelME | Score33.75% | Percentile41.67% | Participants25 | EvidenceC | Evaluated |
| Rank16 | ModelZA | Score33.30% | Percentile37.50% | Participants25 | EvidenceC | Evaluated |
| Rank17 | ModelAMNova 2 LiteAmazon | Score32.50% | Percentile33.33% | Participants25 | EvidenceC | Evaluated |
| Rank18 | ModelDE | Score31.30% | Percentile29.17% | Participants25 | EvidenceC | Evaluated |
| Rank19 | ModelXI | Score30.50% | Percentile25.00% | Participants25 | EvidenceC | Evaluated |
| Rank20 | ModelZA | Score30.00% | Percentile20.83% | Participants25 | EvidenceC | Evaluated |
| Rank21 | ModelMA | Score30.00% | Percentile16.67% | Participants25 | EvidenceC | Evaluated |
| Rank22 | ModelNV | Score25.78% | Percentile12.50% | Participants25 | EvidenceC | Evaluated |
| Rank23 | ModelMA | Score25.00% | Percentile8.33% | Participants25 | EvidenceC | Evaluated |
| Rank24 | ModelNV | Score8.50% | Percentile4.17% | Participants25 | EvidenceC | Evaluated |
| Rank25 | ModelDE | Score5.70% | Percentile0.00% | Participants25 | EvidenceC | Evaluated |
Terminal-Bench Highlights
The leading models and scores on this benchmark.
Terminal-Bench Score Distribution
A closer view of the leading scores on this benchmark.
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.
What is Terminal-Bench?
What Terminal-Bench measures and how its scores work.
Terminal-Bench is a benchmark consisting of a dataset of ~100 hand-crafted, human-verified tasks and an execution harness that connects language models to a terminal sandbox.
It measures how well AI agents handle tasks including compiling code, training models, setting up servers, system administration, security tasks, data science workflows, and cybersecurity vulnerabilities, reported as a Score ratio.
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
- No
- Evaluation key
- overall
LLM Stats. 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 50.00%.
What are the top three models on Terminal-Bench?
The current leaders are Claude Sonnet 4.5 (50.00%), MiniMax M2.1 (47.90%), and Kimi K2-Thinking-0905 (47.10%).
Which Terminal-Bench model has the lowest official input price?
MiMo-V2-Flash has the lowest matched official input price at $0.14 input / $0.28 output per 1M tokens.
Which models are fastest among Terminal-Bench results?
The fastest matched records are MiniMax M2.1 (100.00 tok/s via MiniMax), Claude Opus 4.1 (100.00 tok/s via Anthropic), and Claude Haiku 4.5 (100.00 tok/s via Vertex AI).
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.