Agentic & Tool UseActive
Trust Score:
93

Terminal-Bench-Science 0.1

Terminal-Bench variant focused on scientific and research workflows executed from the command line.

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
Qwen3.8-Flash-Next7.1%2026-11-15independentSource ↗
Gemini 4 Argon52.9%2026-11-15independentSource ↗
Nanbeige4.3-33B1%2026-11-15independentSource ↗
Ling 3.1 Flash 10194.3%2026-11-15independentSource ↗
K2 Horizon MoVA 36B A4B0%2026-11-15independentSource ↗
DeepSeek V4.1 Flash 101712.9%2026-11-15independentSource ↗
Muse Glimmer0%2026-11-15independentSource ↗
Nemotron 3 Ultra0%2026-11-15independentSource ↗
Kimi K3.1 Pro34.3%2026-11-15independentSource ↗
Beam 1.00.5%2026-11-15independentSource ↗
Inkling Small0%2026-11-15independentSource ↗
Gemini 3.8 Flash10%2026-11-15independentSource ↗
Muse Ember20%2026-11-15independentSource ↗
A.X-K20%2026-11-15independentSource ↗
DeepSeek V4-Pro5.7%2026-11-15independentSource ↗
Gemini 4 Hydrogen49.5%2026-11-15independentSource ↗
Muse Wonder25.7%2026-11-15independentSource ↗
NVIDIA Nemotron 3 Nano 4B0%2026-11-15independentSource ↗
Granite 4.2 8B0%2026-11-15independentSource ↗
Qwen3.8 Max (0902)11.9%2026-11-15independentSource ↗
JT-4.1 Flash 236B A21B0%2026-11-15independentSource ↗
K2 Horizon 375B A23B0%2026-11-15independentSource ↗
gpt-oss-120b0%2026-11-15independentSource ↗
GLM 5.39.5%2026-11-15independentSource ↗
Gemma 4 31B0%2026-11-15independentSource ↗
Hy412.9%2026-11-15independentSource ↗
MiniMax M3.1 Flash3.3%2026-11-15independentSource ↗
Agnes 2.5 Pro Theta3.8%2026-11-15independentSource ↗
Mistral Medium 3.50%2026-11-15independentSource ↗
Nemotron 3.5 Super VL0%2026-11-15independentSource ↗
Inkling 2 Small5.7%2026-11-15independentSource ↗
Muse Wonder Full37.1%2026-11-15independentSource ↗
GLM 5.3 Flash4.8%2026-11-15independentSource ↗
Apodex 1.10.5%2026-11-15independentSource ↗
GPT-6.1 Sol58.1%2026-11-15independentSource ↗
Gemini 3.5 Flash-Lite0%2026-11-15independentSource ↗
Step 52.4%2026-11-15independentSource ↗
Granite 4.2 3B0%2026-11-15independentSource ↗
Claude Fable 5.273.3%2026-11-15independentSource ↗
GPT-6.1 Terra57.6%2026-11-15independentSource ↗
Gemini 4 Carbon64.8%2026-11-15independentSource ↗
Inkling0%2026-11-15independentSource ↗
Granite 4.2 30B0%2026-11-15independentSource ↗
K2 Horizon 7B0%2026-11-15independentSource ↗
MiMo-V2.6-Flash4.3%2026-11-15independentSource ↗
Mistral Small 40%2026-11-15independentSource ↗
Solar Pro 40%2026-11-15independentSource ↗
K-EXAONE 2.00%2026-11-15independentSource ↗
Claude Sonnet 5.553.3%2026-11-15independentSource ↗
GPT-6 Luna8.6%2026-11-15independentSource ↗
North Mini Code0%2026-11-15independentSource ↗
GPT-6 Astra63.3%2026-11-15independentSource ↗
Claude Haiku 5.520%2026-11-15independentSource ↗
Grok 4.711.4%2026-11-15independentSource ↗
MiMo-V2.6-Pro5.7%2026-11-15independentSource ↗
MiniMax M30.5%2026-11-15independentSource ↗
Mistral Large 44.3%2026-11-15independentSource ↗
Nemotron 3.5 Lightning0%2026-11-15independentSource ↗
Command A+0%2026-11-15independentSource ↗
Claude Opus 5.559%2026-11-15independentSource ↗
K2 Horizon 3.7B0%2026-11-15independentSource ↗
Qwen3.8 27B0.5%2026-11-15independentSource ↗
Muse Spark Open1.9%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-science --batch_size auto
Option BOpenCompass Evaluation Framework
opencompass --datasets terminal-bench-science --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-Science 0.1, 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