New: Agent Index for knowledge work is live.Explore the index ↗
Compare models

Best LLMs for terminal use

Terminal-Bench

Terminal-Bench evaluates agents on challenging tasks performed through a terminal. It tests multi-step command-line work, problem solving, and the ability to use tools to complete practical tasks. Reported scores depend on the benchmark release and agent harness.

Updated: Data as of:
Widely usedNot saturated
Official benchmark
Terminal-Bench results. Each row identifies its benchmark version and evaluation track. Scores from different versions are not directly comparable. Best effort from the latest stored report date per model and version.
Model / creatorVersionReported score
Kimi K3Moonshot AI · Open source / open weights2.185.019%
GLM-5.3-FlashZ.ai · Open source / open weights2.184.3%
GLM-5.3-FlashZ.ai · Open source / open weights2.184.27%
GLM-5.3Z.ai · Open source / open weights2.183.895%
GLM-5.2Z.ai · Open source / open weights2.181%
DeepSeek V4 Pro 0813DeepSeek · Open source / open weights2.178.652%
GLM-5.2Z.ai · Open source / open weights2.177.903%
DeepSeek V4 Pro 0424DeepSeek · Open source / open weights2.067.9%
Kimi K3.1 ProMoonshot AI · Open source / open weights4.0 / mini-swe-agent67.172%
Kimi K2.6Moonshot AI · Open source / open weights2.066.7%
MiniMax M3MiniMax · Open source / open weights2.166%
Kimi K2.6Moonshot AI · Open source / open weights2.165.918%
MiniMax M3MiniMax · Open source / open weights2.165.169%
Gemini 4 CarbonGoogle · Proprietary4.0 / mini-swe-agent65.152%
Claude Opus 5.5Anthropic · Proprietary4.064.8%
DeepSeek V4 Pro 0424DeepSeek · Open source / open weights2.164.794%
Claude Sonnet 5.5Anthropic · Proprietary4.0 / mini-swe-agent63.636%
Claude Sonnet 5.5Anthropic · Proprietary4.061.8%
DeepSeek V4 Flash 0420DeepSeek · Open source / open weights2.161.798%
Claude Opus 5.5Anthropic · Proprietary4.0 / mini-swe-agent59.596%
GPT-6 AstraOpenAI · Proprietary4.0 / mini-swe-agent59.596%
GPT-6 AstraOpenAI · Proprietary4.058.2%
GPT-6.1 SolOpenAI · Proprietary4.058.2%
Claude Fable 5.1Anthropic · Proprietary4.057.9%
Gemini 4 ArgonGoogle · Proprietary4.0 / mini-swe-agent57.071%
DeepSeek V4 Flash 0420DeepSeek · Open source / open weights2.056.9%
GPT-6.1 SolOpenAI · Proprietary4.0 / mini-swe-agent56.061%
GPT-6.1 TerraOpenAI · Proprietary4.0 / mini-swe-agent55.6%
Claude Fable 5.1Anthropic · Proprietary4.0 / mini-swe-agent55.051%
Claude Opus 5Anthropic · Proprietary4.053.9%
Muse Wonder 1.0Meta · Proprietary4.0 / mini-swe-agent52%
Kimi K2.5Moonshot AI · Open source / open weights2.050.8%
Claude Opus 5Anthropic · Proprietary4.0 / mini-swe-agent48.99%
Kimi K2.5Moonshot AI · Open source / open weights2.145.693%
Claude Fable 5Anthropic · Proprietary4.044.5%
Claude Fable 5Anthropic · Proprietary4.0 / mini-swe-agent42.424%
GLM-5.3Z.ai · Open source / open weights4.0 / mini-swe-agent41.919%
GLM-5.3Z.ai · Open source / open weights4.041.8%
GPT-5.6 SolOpenAI · Proprietary4.0 / mini-swe-agent39.899%
Grok 4.7xAI · Proprietary4.037.6%
GPT-5.6 SolOpenAI · Proprietary4.037.3%
GPT-5.6 TerraOpenAI · Proprietary4.0 / mini-swe-agent35.4%
MiMo-V2.6-ProXiaomi · Open source / open weights4.0 / mini-swe-agent34.8%
DeepSeek V4.1 Flash 1017DeepSeek · Open source / open weights4.0 / mini-swe-agent33.838%
Muse Spark 1.3Meta · Proprietary4.0 / mini-swe-agent33.333%
Step 5 PreviewStepFun · Open source / open weights4.0 / mini-swe-agent33.3%
GLM-5.3-FlashZ.ai · Open source / open weights4.0 / mini-swe-agent32.828%
Claude Haiku 5.5Anthropic · Proprietary4.0 / mini-swe-agent32.8%
DeepSeek V4.1 FlashDeepSeek · Open source / open weights4.0 / mini-swe-agent26.768%
gpt-oss-120bOpenAI · Open source / open weights2.126.217%
Grok 4.7xAI · Proprietary4.0 / mini-swe-agent25.758%
MiniMax M3.1 FlashMiniMax · Open source / open weights4.0 / mini-swe-agent25.758%
Claude Opus 4.8Anthropic · Proprietary4.023.6%
Claude Opus 4.8Anthropic · Proprietary4.0 / mini-swe-agent21.717%
Gemini 3.8 FlashGoogle · Proprietary4.0 / mini-swe-agent19.697%
DeepSeek R1 (Jan 2025)DeepSeek · Open source / open weights2.119.101%
Gemini 3.8 FlashGoogle · Proprietary4.019.1%
DeepSeek V4 Pro 0424DeepSeek · Open source / open weights4.0 / mini-swe-agent14.646%
GPT-5.5OpenAI · Proprietary4.0 / mini-swe-agent14.646%
Muse Spark 1.3Meta · Proprietary4.014.5%
Claude Sonnet 5Anthropic · Proprietary4.0 / mini-swe-agent14.141%
DeepSeek V4 Pro 0813DeepSeek · Open source / open weights4.0 / mini-swe-agent14.141%
gpt-oss-20bOpenAI · Open source / open weights2.113.858%
Kimi K3Moonshot AI · Open source / open weights4.0 / mini-swe-agent12.626%
GPT-6 LunaOpenAI · Proprietary4.0 / mini-swe-agent12.6%
Claude Sonnet 5Anthropic · Proprietary4.012.4%
GPT-5.6 LunaOpenAI · Proprietary4.0 / mini-swe-agent11.6%
Muse Spark OpenMeta · Open source / open weights4.0 / mini-swe-agent8.1%
Muse Spark 1.2Meta · Proprietary4.0 / mini-swe-agent7.071%
Muse Spark 1.1Meta · Proprietary4.0 / mini-swe-agent6.061%
Qwen3.8 27BQwen · Open source / open weights4.0 / mini-swe-agent5.6%
Gemini 3.1 ProGoogle · Proprietary4.0 / mini-swe-agent4.04%
DeepSeek V4 Flash 0420DeepSeek · Open source / open weights4.0 / mini-swe-agent3.03%
Claude Sonnet 4.6Anthropic · Proprietary4.0 / mini-swe-agent3%
DeepSeek V4 FlashDeepSeek · Open source / open weights4.0 / mini-swe-agent3%
MiniMax M3MiniMax · Open source / open weights4.0 / mini-swe-agent2.02%
GLM-5.1Z.ai · Open source / open weights4.0 / mini-swe-agent2%
GPT-5.4 miniOpenAI · Proprietary4.0 / mini-swe-agent2%
GLM-5.2Z.ai · Open source / open weights4.0 / mini-swe-agent1.01%
InklingThinking Machines · Open source / open weights4.0 / mini-swe-agent1%
Kimi K2.6Moonshot AI · Open source / open weights4.0 / mini-swe-agent0.505%
Qwen 3.5 9BQwen · Open source / open weights4.0 / mini-swe-agent0.5%
Claude Sonnet 4.5Anthropic · Proprietary4.0 / mini-swe-agent0%
DeepSeek R1 (Jan 2025)DeepSeek · Open source / open weights4.0 / mini-swe-agent0%
gpt-oss-120bOpenAI · Open source / open weights4.0 / mini-swe-agent0%
gpt-oss-20bOpenAI · Open source / open weights4.0 / mini-swe-agent0%
MiniMax M2.7MiniMax · Open source / open weights4.0 / mini-swe-agent0%
Nemotron 3.5 Super VLNVIDIA · Open source / open weights4.0 / mini-swe-agent0%
Qwen3-14BQwen · Open source / open weights4.0 / mini-swe-agent0%
Qwen3.5-397B-A17BQwen · Open source / open weights4.0 / mini-swe-agent0%
View all 100 published reports
4.0 / mini-swe-agent

harness: mini-swe-agent · repeats: 3 · selection: best reported configuration per release · Kimi K3.1 Pro

4.0 / mini-swe-agent

harness: mini-swe-agent · repeats: 3 · selection: best reported configuration per release · Gemini 4 Carbon (High)

4.0 / mini-swe-agent

harness: mini-swe-agent · repeats: 3 · selection: best reported configuration per release · Claude Sonnet 5.5 (Max, Default Fallback)

4.0 / mini-swe-agent

harness: mini-swe-agent · repeats: 3 · selection: best reported configuration per release · Claude Opus 5.5 (Max, Default Fallback)

4.0 / mini-swe-agent

harness: mini-swe-agent · repeats: 3 · selection: best reported configuration per release · GPT-6 Astra (Xhigh)

4.0 / mini-swe-agent

harness: mini-swe-agent · repeats: 3 · selection: best reported configuration per release · Gemini 4 Argon (High)

4.0 / mini-swe-agent

harness: mini-swe-agent · repeats: 3 · selection: best reported configuration per release · GPT-6.1 Sol (Max)

4.0 / mini-swe-agent

harness: mini-swe-agent · repeats: 3 · selection: best reported configuration per release · GPT-6.1 Terra (max)

4.0 / mini-swe-agent

harness: mini-swe-agent · repeats: 3 · selection: best reported configuration per release · Claude Fable 5.1 (Xhigh, Default Fallback)

4.0 / mini-swe-agent

harness: mini-swe-agent · repeats: 3 · selection: best reported configuration per release · Muse Wonder (max)

4.0 / mini-swe-agent

harness: mini-swe-agent · repeats: 3 · selection: best reported configuration per release · Claude Opus 5 (Max)

4.0 / mini-swe-agent

harness: mini-swe-agent · repeats: 3 · selection: best reported configuration per release · Claude Fable 5 (Max, Opus 4.8 Fallback)

GLM-5.341.919%
4.0 / mini-swe-agent

harness: mini-swe-agent · repeats: 3 · selection: best reported configuration per release · GLM-5.3 (Max)

4.0 / mini-swe-agent

harness: mini-swe-agent · repeats: 3 · selection: best reported configuration per release · GPT-5.6 Sol (Max)

4.0 / mini-swe-agent

harness: mini-swe-agent · repeats: 3 · selection: best reported configuration per release · GPT-5.6 Terra (max)

4.0 / mini-swe-agent

harness: mini-swe-agent · repeats: 3 · selection: best reported configuration per release · MiMo-V2.6-Pro

4.0 / mini-swe-agent

harness: mini-swe-agent · repeats: 3 · selection: best reported configuration per release · DeepSeek V4.1 Flash 1017 (Max)

4.0 / mini-swe-agent

harness: mini-swe-agent · repeats: 3 · selection: best reported configuration per release · Muse Spark 1.3 (Max)

4.0 / mini-swe-agent

harness: mini-swe-agent · repeats: 3 · selection: best reported configuration per release · Step 5 Preview

4.0 / mini-swe-agent

harness: mini-swe-agent · repeats: 3 · selection: best reported configuration per release · GLM-5.3-Flash

4.0 / mini-swe-agent

harness: mini-swe-agent · repeats: 3 · selection: best reported configuration per release · Claude Haiku 5.5 (max)

4.0 / mini-swe-agent

harness: mini-swe-agent · repeats: 3 · selection: best reported configuration per release · DeepSeek V4.1 Flash (Max)

4.0 / mini-swe-agent

harness: mini-swe-agent · repeats: 3 · selection: best reported configuration per release · MiniMax-M3.1-Flash

Grok 4.725.758%
4.0 / mini-swe-agent

harness: mini-swe-agent · repeats: 3 · selection: best reported configuration per release · Grok 4.7 (Xhigh)

4.0 / mini-swe-agent

harness: mini-swe-agent · repeats: 3 · selection: best reported configuration per release · Claude Opus 4.8 (Max)

4.0 / mini-swe-agent

harness: mini-swe-agent · repeats: 3 · selection: best reported configuration per release · Gemini 3.8 Flash (Medium)

4.0 / mini-swe-agent

harness: mini-swe-agent · repeats: 3 · selection: best reported configuration per release · DeepSeek V4 Pro 0424 (Max)

GPT-5.514.646%
4.0 / mini-swe-agent

harness: mini-swe-agent · repeats: 3 · selection: best reported configuration per release · GPT-5.5 (Xhigh)

4.0 / mini-swe-agent

harness: mini-swe-agent · repeats: 3 · selection: best reported configuration per release · DeepSeek V4 Pro 0813 (Max)

4.0 / mini-swe-agent

harness: mini-swe-agent · repeats: 3 · selection: best reported configuration per release · Claude Sonnet 5 (Max)

Kimi K312.626%
4.0 / mini-swe-agent

harness: mini-swe-agent · repeats: 3 · selection: best reported configuration per release · Kimi K3 (Low)

4.0 / mini-swe-agent

harness: mini-swe-agent · repeats: 3 · selection: best reported configuration per release · GPT-6 Luna (max)

4.0 / mini-swe-agent

harness: mini-swe-agent · repeats: 3 · selection: best reported configuration per release · GPT-5.6 Luna (max)

4.0 / mini-swe-agent

harness: mini-swe-agent · repeats: 3 · selection: best reported configuration per release · Muse Spark Open (xhigh)

4.0 / mini-swe-agent

harness: mini-swe-agent · repeats: 3 · selection: best reported configuration per release · Muse Spark 1.2 (Xhigh)

4.0 / mini-swe-agent

harness: mini-swe-agent · repeats: 3 · selection: best reported configuration per release · Muse Spark 1.1 (Xhigh)

4.0 / mini-swe-agent

harness: mini-swe-agent · repeats: 3 · selection: best reported configuration per release · Qwen3.8 27B (xhigh)

4.0 / mini-swe-agent

harness: mini-swe-agent · repeats: 3 · selection: best reported configuration per release · Gemini 3.1 Pro Preview

4.0 / mini-swe-agent

harness: mini-swe-agent · repeats: 3 · selection: best reported configuration per release · DeepSeek V4 Flash 0420 (High)

4.0 / mini-swe-agent

harness: mini-swe-agent · repeats: 3 · selection: best reported configuration per release · DeepSeek V4 Flash (high)

4.0 / mini-swe-agent

harness: mini-swe-agent · repeats: 3 · selection: best reported configuration per release · Claude Sonnet 4.6 (max)

4.0 / mini-swe-agent

harness: mini-swe-agent · repeats: 3 · selection: best reported configuration per release · MiniMax-M3

4.0 / mini-swe-agent

harness: mini-swe-agent · repeats: 3 · selection: best reported configuration per release · GLM-5.1

4.0 / mini-swe-agent

harness: mini-swe-agent · repeats: 3 · selection: best reported configuration per release · GPT-5.4 mini (xhigh)

GLM-5.21.01%
4.0 / mini-swe-agent

harness: mini-swe-agent · repeats: 3 · selection: best reported configuration per release · GLM-5.2 (Max)

4.0 / mini-swe-agent

harness: mini-swe-agent · repeats: 3 · selection: best reported configuration per release · Inkling (xhigh)

Kimi K2.60.505%
4.0 / mini-swe-agent

harness: mini-swe-agent · repeats: 3 · selection: best reported configuration per release · Kimi K2.6 (Reasoning)

4.0 / mini-swe-agent

harness: mini-swe-agent · repeats: 3 · selection: best reported configuration per release · Qwen3.5 9B

4.0 / mini-swe-agent

harness: mini-swe-agent · repeats: 3 · selection: best reported configuration per release · Nemotron 3.5 Super VL

4.0 / mini-swe-agent

harness: mini-swe-agent · repeats: 3 · selection: best reported configuration per release · MiniMax-M2.7

4.0 / mini-swe-agent

harness: mini-swe-agent · repeats: 3 · selection: best reported configuration per release · Qwen3 14B

4.0 / mini-swe-agent

harness: mini-swe-agent · repeats: 3 · selection: best reported configuration per release · Qwen3.5 397B A17B

4.0 / mini-swe-agent

harness: mini-swe-agent · repeats: 3 · selection: best reported configuration per release · Claude 4.5 Sonnet

4.0 / mini-swe-agent

harness: mini-swe-agent · repeats: 3 · selection: best reported configuration per release · gpt-oss-120b (High)

4.0 / mini-swe-agent

harness: mini-swe-agent · repeats: 3 · selection: best reported configuration per release · gpt-oss-20b (High)

4.0 / mini-swe-agent

harness: mini-swe-agent · repeats: 3 · selection: best reported configuration per release · DeepSeek R1 (Jan '25)

2.1

harness comparison: provider-reported; settings differ across models · Claude Code 2.1.207; 6h; 65536 output tokens

Z.ai ↗
2.1

harness comparison: provider-reported; settings differ across models · Terminus 2; JSON parser; 4h; 500 episodes; 256K context

Z.ai ↗
2.1

harness comparison: provider-reported; settings differ across models · Terminus 2; 8 CPU / 16GB; 2h; 128K output

MiniMax ↗
Kimi K385.019%
2.1

Kimi K3 (Max); 89 tasks; Terminus 2; e2b; three repeats

2.1

GLM-5.3-Flash; 89 tasks; Terminus 2; e2b; three repeats

GLM-5.383.895%
2.1

GLM-5.3 (Max); 89 tasks; Terminus 2; e2b; three repeats

Kimi K382.397%
2.1

Kimi K3 (Low); 89 tasks; Terminus 2; e2b; three repeats

GLM-5.277.903%
2.1

GLM-5.2 (Max); 89 tasks; Terminus 2; e2b; three repeats

Kimi K2.665.918%
2.1

Kimi K2.6 (Reasoning); 89 tasks; Terminus 2; e2b; three repeats

MiniMax M365.169%
2.1

MiniMax-M3; 89 tasks; Terminus 2; e2b; three repeats

GLM-5.251.685%
2.1

GLM-5.2 (Non-reasoning); 89 tasks; Terminus 2; e2b; three repeats

Kimi K2.545.693%
2.1

Kimi K2.5 (Reasoning); 89 tasks; Terminus 2; e2b; three repeats

2.1

gpt-oss-120b (High); 89 tasks; Terminus 2; e2b; three repeats

2.1

gpt-oss-120b (Low); 89 tasks; Terminus 2; e2b; three repeats

2.1

gpt-oss-20b (High); 89 tasks; Terminus 2; e2b; three repeats

2.0

harness comparison: provider-reported; settings differ across models · Published Terminal-Bench 2.0 score; not 2.1

Moonshot AI ↗
2.0

harness comparison: provider-reported; settings differ across models · Published Terminal-Bench 2.0 score; not 2.1

Moonshot AI ↗