reasoning benchmark
Terminal-Bench 2.0 is an updated benchmark for evaluating AI agents' ability to use tools and operate a computer through a terminal.
Updated Sep 5, 2026
Higher score ranks better on this benchmark.
Rank | Model | Score | Percentile | Participants | Evidence | Evaluated |
|---|
| Rank01 | ModelOP | Score82.70% | Percentile100.00% | Participants51 | EvidenceC | Evaluated |
| Rank02 | ModelAN | Score82.00% | Percentile98.00% | Participants51 | EvidenceC | Evaluated |
| Rank03 | ModelAN | Score80.40% | Percentile96.00% | Participants51 | EvidenceC | Evaluated |
| Rank04 | ModelOP | Score77.30% | Percentile94.00% | Participants51 | EvidenceC | Evaluated |
| Rank05 | ModelGO | Score76.20% | Percentile92.00% | Participants51 | EvidenceC | Evaluated |
| Rank06 | ModelOP | Score75.10% | Percentile90.00% | Participants51 | EvidenceC | Evaluated |
| Rank07 | ModelAN | Score74.60% | Percentile88.00% | Participants51 | EvidenceC | Evaluated |
| Rank08 | ModelAC | Score70.30% | Percentile86.00% | Participants51 | EvidenceC | Evaluated |
| Rank09 | ModelAC | Score69.70% | Percentile84.00% | Participants51 | EvidenceC | Evaluated |
| Rank10 | ModelAN | Score69.40% | Percentile82.00% | Participants51 | EvidenceC | Evaluated |
| Rank11 | ModelZA | Score69.00% | Percentile80.00% | Participants51 | EvidenceC | Evaluated |
| Rank12 | ModelGO | Score68.50% | Percentile78.00% | Participants51 | EvidenceC | Evaluated |
| Rank13 | ModelXI | Score68.40% | Percentile76.00% | Participants51 | EvidenceC | Evaluated |
| Rank14 | ModelDE | Score67.90% | Percentile74.00% | Participants51 | EvidenceC | Evaluated |
| Rank15 | ModelMA | Score66.70% | Percentile72.00% | Participants51 | EvidenceC | Evaluated |
| Rank16 | ModelXI | Score65.80% | Percentile70.00% | Participants51 | EvidenceC | Evaluated |
| Rank17 | ModelAN | Score65.40% | Percentile68.00% | Participants51 | EvidenceC | Evaluated |
| Rank18 | ModelOP | Score64.00% | Percentile66.00% | Participants51 | EvidenceC | Evaluated |
| Rank19 | ModelAC | Score61.60% | Percentile64.00% | Participants51 | EvidenceC | Evaluated |
| Rank20 | ModelOP | Score60.00% | Percentile62.00% | Participants51 | EvidenceC | Evaluated |
| Rank21 | ModelAN | Score59.30% | Percentile60.00% | Participants51 | EvidenceC | Evaluated |
| Rank22 | ModelAC | Score59.30% | Percentile58.00% | Participants51 | EvidenceC | Evaluated |
| Rank23 | ModelAN | Score59.10% | Percentile56.00% | Participants51 | EvidenceC | Evaluated |
| Rank24 | ModelME | Score59.00% | Percentile54.00% | Participants51 | EvidenceC | Evaluated |
| Rank25 | ModelXI | Score57.10% | Percentile52.00% | Participants51 | EvidenceC | Evaluated |
| Rank26 | ModelMI | Score57.00% | Percentile50.00% | Participants51 | EvidenceC | Evaluated |
| Rank27 | ModelDE | Score56.90% | Percentile48.00% | Participants51 | EvidenceC | Evaluated |
| Rank28 | ModelDE | Score56.60% | Percentile46.00% | Participants51 | EvidenceC | Evaluated |
| Rank29 | ModelZA | Score56.20% | Percentile44.00% | Participants51 | EvidenceC | Evaluated |
| Rank30 | ModelMI | Score54.80% | Percentile42.00% | Participants51 | EvidenceC | Evaluated |
The leading models and scores on this benchmark.
A closer view of the leading scores on this benchmark.
The first five results on this benchmark, with official price and output speed added where the model identity can be matched.
What Terminal-Bench 2.0 measures and how its scores work.
Terminal-Bench 2.0 is a benchmark for testing AI agents on real-world, end-to-end tasks performed autonomously through a terminal.
It measures how well models handle tasks including compiling code, training models, setting up servers, system administration, security tasks, data science workflows, and cybersecurity vulnerabilities, reported as Score in ratio units.
Scores are shown in ratio. This benchmark is not independently verified and has an evidence level of C.
LLM Stats. Benchmark scores retain their original unit. Overall score eligibility is shown separately.
Common questions about Terminal-Bench 2.0.
GPT-5.5 is currently ranked first with 82.70%.
The current leaders are GPT-5.5 (82.70%), Claude Mythos Preview (82.00%), and Claude Sonnet 5 (80.40%).
Step-3.5-Flash has the lowest matched official input price at $0.10 input / $0.30 output per 1M tokens.
The fastest matched records are Gemini 3.5 Flash (210.94 tok/s via Google), GPT-5.5 (134.94 tok/s via OpenAI), and Gemini 3 Flash (124.32 tok/s via Google).
No. This benchmark measures one defined capability or task. The overall LLMBoard score uses a separate aggregation across eligible benchmark evidence.
It measures how well models handle tasks including compiling code, training models, setting up servers, system administration, security tasks, data science workflows, and cybersecurity vulnerabilities, reported as Score in ratio units.
Yes. Higher values rank better for this benchmark.
51 model results are currently shown.
No. This benchmark is shown for reference but does not contribute to the overall score.
Ranking basisThis terminal-bench 2.0 AI model leaderboard uses descending score in the benchmark original unit. The leaderboard ranking keeps matched price and speed data separate from benchmark evidence.
Selection summary
GPT-5.5 currently leads Terminal-Bench 2.0 with 82.70%. 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.