Agentic & Tool UseActive
Trust Score:
93

Terminal-Bench Hard

Hard subset of Terminal-Bench: autonomous agents completing real tasks in a Linux terminal, as run by Artificial Analysis.

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
JT-35B-Flash28.8%2026-11-15independentSource ↗
Qwen3.5 Omni Plus21.2%2026-11-15independentSource ↗
Nova 2.0 Omni4.5%2026-11-15independentSource ↗
Qwen3.5-122B-A10B29.5%2026-11-15independentSource ↗
Gemma 4 12B11.4%2026-11-15independentSource ↗
Gemma 3 270M0%2026-11-15independentSource ↗
Qwen3.5 2B3.8%2026-11-15independentSource ↗
Qwen3 Omni 30B A3B Instruct1.5%2026-11-15independentSource ↗
Nova 2.0 Pro24.2%2026-11-15independentSource ↗
Qwen3.5 397B A17B40.9%2026-11-15independentSource ↗
Mistral Small 417.4%2026-11-15independentSource ↗
gpt-oss-20b4.5%2026-11-15independentSource ↗
o337.1%2026-11-15independentSource ↗
North Mini Code31.1%2026-11-15independentSource ↗
Granite 4.0 Micro1.5%2026-11-15independentSource ↗
Qwen3.5-9B18.2%2026-11-15independentSource ↗
Hermes 4 - Llama-3.1 70B0%2026-11-15independentSource ↗
Nemotron 3 Nano 30B A3B12.1%2026-11-15independentSource ↗
Qwen3 Next 80B A3B Instruct7.6%2026-11-15independentSource ↗
MiniMax M342.4%2026-11-15independentSource ↗
NVIDIA Nemotron Nano 9B V20.8%2026-11-15independentSource ↗
Command A+25%2026-11-15independentSource ↗
EXAONE 4.0 32B1.5%2026-11-15independentSource ↗
Qwen3.5 4B11.4%2026-11-15independentSource ↗
Olmo 3 7B Think0.8%2026-11-15independentSource ↗
HyperCLOVA X SEED Think (32B)12.1%2026-11-15independentSource ↗
HyperNova 60B 260523.5%2026-11-15independentSource ↗
Nova Micro1.5%2026-11-15independentSource ↗
INTELLECT-39.1%2026-11-15independentSource ↗
Olmo 3.1 32B Think0%2026-11-15independentSource ↗
LFM2.5-VL-1.6B0%2026-11-15independentSource ↗
JT-MINI18.2%2026-11-15independentSource ↗
Molmo2-8B0%2026-11-15independentSource ↗
Qwen3.5 0.8B0%2026-11-15independentSource ↗
MiniCPM-V 4.6 1.3B0%2026-11-15independentSource ↗
Sarvam 105B1.5%2026-11-15independentSource ↗
Llama 3.1 Instruct 405B6.8%2026-11-15independentSource ↗
Ring-flash-2.07.6%2026-11-15independentSource ↗
LFM2.5-1.2B-Instruct0%2026-11-15independentSource ↗
Apertus 70B Instruct0%2026-11-15independentSource ↗
Doubao Seed Code26.5%2026-11-15independentSource ↗
Gemma 4 E4B7.6%2026-11-15independentSource ↗
Granite 4.0 350M0%2026-11-15independentSource ↗
Kimi Linear 48B A3B Instruct11.4%2026-11-15independentSource ↗
Solar Open 100B2.3%2026-11-15independentSource ↗
GPT-5.3-Codex53%2026-11-15independentSource ↗
LFM2.5-1.2B-Thinking0%2026-11-15independentSource ↗
K2 Think V26.8%2026-11-15independentSource ↗
Nemotron 3 Ultra36.4%2026-11-15independentSource ↗
Tri-21B-Think0.8%2026-11-15independentSource ↗
Granite 4.0 H Small2.3%2026-11-15independentSource ↗
Llama 3.2 Instruct 11B (Vision)0.8%2026-11-15independentSource ↗
LongCat Flash Lite10.6%2026-11-15independentSource ↗
Nova 2.0 Lite16.7%2026-11-15independentSource ↗
Motif-2-12.7B-Reasoning3.8%2026-11-15independentSource ↗
Qwen3.6 35B A3B25.8%2026-11-15independentSource ↗
Ministral 3 3B0%2026-11-15independentSource ↗
Gemma 4 26B A4B13.6%2026-11-15independentSource ↗
NVIDIA Nemotron Nano 12B v2 VL4.5%2026-11-15independentSource ↗
Qwen3.5 Omni Flash8.3%2026-11-15independentSource ↗
Magistral Small 1.24.5%2026-11-15independentSource ↗
Ministral 3 14B4.5%2026-11-15independentSource ↗
Gemma 4 E2B2.3%2026-11-15independentSource ↗
Sarvam 30B2.3%2026-11-15independentSource ↗
MiniCPM5-1B0%2026-11-15independentSource ↗
Jamba 1.7 Large2.3%2026-11-15independentSource ↗
Nanbeige4.1-3B0%2026-11-15independentSource ↗
Tiny Aya Global0%2026-11-15independentSource ↗
Exaone 4.0 1.2B0%2026-11-15independentSource ↗
Hermes 4 - Llama-3.1 405B11.4%2026-11-15independentSource ↗
Magistral Medium 1.212.9%2026-11-15independentSource ↗
Apertus 8B Instruct0%2026-11-15independentSource ↗
Nemotron 3 Nano Omni 30B A3B Reasoning8.3%2026-11-15independentSource ↗
Kimi K2.7 Code44.7%2026-11-15independentSource ↗
Olmo 3 7B Instruct0%2026-11-15independentSource ↗
Llama 4 Maverick6.8%2026-11-15independentSource ↗
Olmo 3.1 32B Instruct0%2026-11-15independentSource ↗
Ministral 3 8B4.5%2026-11-15independentSource ↗
Llama 3.3 Instruct 70B3%2026-11-15independentSource ↗
Nemotron Cascade 2 30B A3B21.2%2026-11-15independentSource ↗
NVIDIA Nemotron 3 Nano 4B6.8%2026-11-15independentSource ↗
Phi-4-mini0%2026-11-15independentSource ↗
K2-V29.8%2026-11-15independentSource ↗
ERNIE 5.0 Thinking25%2026-11-15independentSource ↗
Granite 4.0 H 350M0%2026-11-15independentSource ↗
Apriel-v1.6-15B-Thinker14.4%2026-11-15independentSource ↗
Qwen3.7 Plus47%2026-11-15independentSource ↗
Step3 VL 10B5.3%2026-11-15independentSource ↗
Llama 4 Scout1.5%2026-11-15independentSource ↗
ERNIE 4.5 300B A47B6.1%2026-11-15independentSource ↗
Molmo 7B-D0%2026-11-15independentSource ↗
Jamba 1.7 Mini0%2026-11-15independentSource ↗
Trinity Large Thinking22.7%2026-11-15independentSource ↗
Qwen3 Omni 30B A3B (Reasoning)3.8%2026-11-15independentSource ↗
Gemini 3.1 Pro53.8%2026-11-15independentSource ↗
Falcon-H1R-7B2.3%2026-11-15independentSource ↗
Qwen3 Next 80B A3B (Reasoning)9.8%2026-11-15independentSource ↗
gpt-oss-120b23.5%2026-11-15independentSource ↗
Gemma 4 31B36.4%2026-11-15independentSource ↗
Llama 3.1 Nemotron Ultra 253B v12.3%2026-11-15independentSource ↗
Granite 4.0 H 1B0%2026-11-15independentSource ↗
Reka Flash 30%2026-11-15independentSource ↗
Mistral Medium 3.533.3%2026-11-15independentSource ↗
Mi:dm K 2.5 Pro2.3%2026-11-15independentSource ↗
LFM2.5-8B-A1B4.5%2026-11-15independentSource ↗
Nova Premier6.8%2026-11-15independentSource ↗
Cogito v2.116.7%2026-11-15independentSource ↗
Ring-2.6-1T28.8%2026-11-15independentSource ↗
Phi-43.8%2026-11-15independentSource ↗
Jamba Reasoning 3B0.8%2026-11-15independentSource ↗
LFM2 24B A2B0%2026-11-15independentSource ↗
KAT Coder Pro V249.2%2026-11-15independentSource ↗
Qwen3 Coder Next18.2%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-hard --batch_size auto
Option BOpenCompass Evaluation Framework
opencompass --datasets terminal-bench-hard --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 Hard, 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