Agentic & Tool UseActive
Trust Score:
93

Terminal-Bench 4.0

Version 4.0 of Terminal-Bench, tracked by Artificial Analysis for terminal-based autonomous agents.

Launched: —Refresh: static
Status Assessment (active):

Actively evaluated by Artificial Analysis on current frontier models; no saturation evidence recorded.

Performance Timeline

Longitudinal progression of model scores against human baselines.

Performance & Historical Trajectory

Empirical score progression across model release dates and evaluation rounds.

Independent Vendor-Reported
ModelScoreDateSource TypeProvenance
Celeris-10%2026-11-15independentSource ↗
Ling 3.1 Flash33.3%2026-11-15independentSource ↗
NVIDIA Nemotron 3 Nano 4B0%2026-11-15independentSource ↗
K2 Horizon 0.9B0%2026-11-15independentSource ↗
Qwen3.8-Flash-Next25.3%2026-11-15independentSource ↗
Ministral 3 8B0%2026-11-15independentSource ↗
Llama 4 Maverick0%2026-11-15independentSource ↗
Qwen3.8 2.4T A95B11.1%2026-11-15independentSource ↗
Kimi K2.7 Code1%2026-11-15independentSource ↗
Gemini 4 Hydrogen53.5%2026-11-15independentSource ↗
DeepSeek V4-Pro14.1%2026-11-15independentSource ↗
A.X-K20%2026-11-15independentSource ↗
Qwen3.7 Plus1%2026-11-15independentSource ↗
Ling 3.0 Tiny0%2026-11-15independentSource ↗
Nemotron 3.5 Bolt 74B-A9B0%2026-11-15independentSource ↗
Muse Ember40.9%2026-11-15independentSource ↗
DeepSeek V4 Flash Vision12.1%2026-11-15independentSource ↗
Mercury 2.50%2026-11-15independentSource ↗
Gemini 3.8 Flash19.7%2026-11-15independentSource ↗
Inkling Small1%2026-11-15independentSource ↗
Ministral 3 14B0%2026-11-15independentSource ↗
Ling-3.0-flash-Fin0%2026-11-15independentSource ↗
Beam 1.03%2026-11-15independentSource ↗
GLM 5.341.9%2026-11-15independentSource ↗
Qwen3 Coder Next0%2026-11-15independentSource ↗
Quasar 438B1%2026-11-15independentSource ↗
Ring-2.6-1T0.5%2026-11-15independentSource ↗
Muse Wonder Full57.1%2026-11-15independentSource ↗
Inkling 2 Small13.1%2026-11-15independentSource ↗
Nemotron 3.5 Super VL0%2026-11-15independentSource ↗
Mistral Medium 3.50%2026-11-15independentSource ↗
Agnes 2.5 Pro Theta10.1%2026-11-15independentSource ↗
MiniMax M3.1 Flash25.8%2026-11-15independentSource ↗
Hy437.4%2026-11-15independentSource ↗
Gemma 4 31B0%2026-11-15independentSource ↗
Ministral 3 3B0%2026-11-15independentSource ↗
gpt-oss-120b0%2026-11-15independentSource ↗
GPT-5.5 Instant12.6%2026-11-15independentSource ↗
Gemini 3.1 Pro4%2026-11-15independentSource ↗
K2 Horizon 375B A23B1.5%2026-11-15independentSource ↗
Trinity Large Thinking0.5%2026-11-15independentSource ↗
JT-4.1 Flash 236B A21B15.7%2026-11-15independentSource ↗
Hy30.5%2026-11-15independentSource ↗
Qwen3.8 Max (0902)38.9%2026-11-15independentSource ↗
Granite 4.2 8B0%2026-11-15independentSource ↗
Llama 4 Scout0%2026-11-15independentSource ↗
Granite 4.2 30B0%2026-11-15independentSource ↗
Grok 4.725.8%2026-11-15independentSource ↗
Claude Haiku 5.532.8%2026-11-15independentSource ↗
GPT-6 Astra59.1%2026-11-15independentSource ↗
North Mini Code0.5%2026-11-15independentSource ↗
GPT-6 Luna12.6%2026-11-15independentSource ↗
GLM 5.3 Flash32.8%2026-11-15independentSource ↗
K-EXAONE 2.00%2026-11-15independentSource ↗
Solar Pro 40%2026-11-15independentSource ↗
Mistral Small 40%2026-11-15independentSource ↗
MiMo-V2.6-Flash22.7%2026-11-15independentSource ↗
K2 Horizon 7B1%2026-11-15independentSource ↗
MiMo-V2.6-Pro34.8%2026-11-15independentSource ↗
Qwen3.5 397B A17B0%2026-11-15independentSource ↗
Inkling1%2026-11-15independentSource ↗
Gemini 4 Carbon65.2%2026-11-15independentSource ↗
GPT-6.1 Terra55.6%2026-11-15independentSource ↗
Claude Fable 5.259.6%2026-11-15independentSource ↗
Granite 4.2 3B0%2026-11-15independentSource ↗
Step 533.3%2026-11-15independentSource ↗
Gemini 3.5 Flash-Lite1%2026-11-15independentSource ↗
GPT-6.1 Sol56.1%2026-11-15independentSource ↗
Apodex 1.10%2026-11-15independentSource ↗
Ling-3.0-flash-VL0%2026-11-15independentSource ↗
Kimi K3.1 Pro67.2%2026-11-15independentSource ↗
MiniCPM5-2B0%2026-11-15independentSource ↗
Solar Mini 41%2026-11-15independentSource ↗
LongCat 2.00%2026-11-15independentSource ↗
Nemotron 3 Ultra0.5%2026-11-15independentSource ↗
Muse Glimmer0.5%2026-11-15independentSource ↗
DeepSeek V4.1 Flash 101733.8%2026-11-15independentSource ↗
K2 Horizon MoVA 36B A4B0%2026-11-15independentSource ↗
Ling 3.1 Flash 101932.8%2026-11-15independentSource ↗
Nanbeige4.3-33B29.8%2026-11-15independentSource ↗
Gemini 4 Argon57.1%2026-11-15independentSource ↗
Claude Sonnet 5.563.6%2026-11-15independentSource ↗
Muse Wonder52%2026-11-15independentSource ↗
Muse Spark Open8.1%2026-11-15independentSource ↗
Qwen3.8 27B5.6%2026-11-15independentSource ↗
HyperNova 60B 26050%2026-11-15independentSource ↗
K2 Horizon 3.7B0%2026-11-15independentSource ↗
Claude Opus 5.559.6%2026-11-15independentSource ↗
Command A+0.5%2026-11-15independentSource ↗
Nemotron 3.5 Lightning0.5%2026-11-15independentSource ↗
Mistral Large 424.7%2026-11-15independentSource ↗
MiniMax M32%2026-11-15independentSource ↗

Human Baseline & Difficulty Horizon

Calibrated human reference points, specialist benchmarks, and ceiling thresholds.
Measured Human Scoren/aNot established
Baseline Protocol & Interpretation

No peer-reviewed human baseline has been published for this benchmark.

Metric & Scoring Methodology

Verification protocols, aggregation formulas, and specialized metric variants.
Primary Metric:task success (%)
Scoring Engine:unit-tests

Dataset & Compute Cost

Evaluation volume, public availability, API pricing, and local hardware requirements.
Total Dataset Size—Annotated evaluation items
Public Test SetPublicOpenly mirrored on repositories
Access GatingOpen AccessUnrestricted download
Evaluation LicenseOpen AccessDataset usage and redistribution terms
Frontier API Compute Cost:

$5 – $20 USD for full benchmark evaluation run on frontier APIs.

Recommended Local GPU Setup:

1x NVIDIA RTX 4090 (24GB) or A100 (40GB/80GB) via vLLM / SGLang

How to Run & Reproduce

Standardized evaluation protocols, CLI commands, and reproducible runner templates.
Prompt Regimezero-shot
Reasoning Modedirect
Sampling Temp0
Pass@k Budgetk = 1
Tools & SandboxPure Text
Scoring Verifierunit-tests
Option AEleutherAI LM-Evaluation-Harness (Open-Weight Models)
lm_eval --model hf --model_args pretrained=<model_path> --tasks terminal-bench-4-0 --batch_size auto
Option BOpenCompass Evaluation Framework
opencompass --datasets terminal-bench-4-0 --models <model_config>
Python APIDeterministic Inference Loop Snippet
# Standard API Evaluation Loop
from openai import OpenAI

client = OpenAI()
response = client.chat.completions.create(
    model="gpt-4o",
    messages=[{"role": "user", "content": prompt}],
    temperature=0.0,
)
Standardized Reporting Requirement:

When publishing results for Terminal-Bench 4.0, always report the exact prompt template, few-shot exemplar ordering, sampling temperature (temperature=0), maximum reasoning budget tokens, and the precise timestamped model snapshot ID.

Contamination & Memorization Analysis

Audit of pretraining exposure risks, memorization vectors, and refresh policies.
Overall Contamination Risk:MEDIUM
Refresh Cadence:

Static fixed snapshot

Test Set Exposure:

Public on web / HuggingFace