Terminal-Bench-Science 0.1
Terminal-Bench variant focused on scientific and research workflows executed from the command line.
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.
| Model | Score | Date | Source Type | Provenance |
|---|---|---|---|---|
| Qwen3.8-Flash-Next | 7.1% | 2026-11-15 | independent | Source ↗ |
| Gemini 4 Argon | 52.9% | 2026-11-15 | independent | Source ↗ |
| Nanbeige4.3-33B | 1% | 2026-11-15 | independent | Source ↗ |
| Ling 3.1 Flash 1019 | 4.3% | 2026-11-15 | independent | Source ↗ |
| K2 Horizon MoVA 36B A4B | 0% | 2026-11-15 | independent | Source ↗ |
| DeepSeek V4.1 Flash 1017 | 12.9% | 2026-11-15 | independent | Source ↗ |
| Muse Glimmer | 0% | 2026-11-15 | independent | Source ↗ |
| Nemotron 3 Ultra | 0% | 2026-11-15 | independent | Source ↗ |
| Kimi K3.1 Pro | 34.3% | 2026-11-15 | independent | Source ↗ |
| Beam 1.0 | 0.5% | 2026-11-15 | independent | Source ↗ |
| Inkling Small | 0% | 2026-11-15 | independent | Source ↗ |
| Gemini 3.8 Flash | 10% | 2026-11-15 | independent | Source ↗ |
| Muse Ember | 20% | 2026-11-15 | independent | Source ↗ |
| A.X-K2 | 0% | 2026-11-15 | independent | Source ↗ |
| DeepSeek V4-Pro | 5.7% | 2026-11-15 | independent | Source ↗ |
| Gemini 4 Hydrogen | 49.5% | 2026-11-15 | independent | Source ↗ |
| Muse Wonder | 25.7% | 2026-11-15 | independent | Source ↗ |
| NVIDIA Nemotron 3 Nano 4B | 0% | 2026-11-15 | independent | Source ↗ |
| Granite 4.2 8B | 0% | 2026-11-15 | independent | Source ↗ |
| Qwen3.8 Max (0902) | 11.9% | 2026-11-15 | independent | Source ↗ |
| JT-4.1 Flash 236B A21B | 0% | 2026-11-15 | independent | Source ↗ |
| K2 Horizon 375B A23B | 0% | 2026-11-15 | independent | Source ↗ |
| gpt-oss-120b | 0% | 2026-11-15 | independent | Source ↗ |
| GLM 5.3 | 9.5% | 2026-11-15 | independent | Source ↗ |
| Gemma 4 31B | 0% | 2026-11-15 | independent | Source ↗ |
| Hy4 | 12.9% | 2026-11-15 | independent | Source ↗ |
| MiniMax M3.1 Flash | 3.3% | 2026-11-15 | independent | Source ↗ |
| Agnes 2.5 Pro Theta | 3.8% | 2026-11-15 | independent | Source ↗ |
| Mistral Medium 3.5 | 0% | 2026-11-15 | independent | Source ↗ |
| Nemotron 3.5 Super VL | 0% | 2026-11-15 | independent | Source ↗ |
| Inkling 2 Small | 5.7% | 2026-11-15 | independent | Source ↗ |
| Muse Wonder Full | 37.1% | 2026-11-15 | independent | Source ↗ |
| GLM 5.3 Flash | 4.8% | 2026-11-15 | independent | Source ↗ |
| Apodex 1.1 | 0.5% | 2026-11-15 | independent | Source ↗ |
| GPT-6.1 Sol | 58.1% | 2026-11-15 | independent | Source ↗ |
| Gemini 3.5 Flash-Lite | 0% | 2026-11-15 | independent | Source ↗ |
| Step 5 | 2.4% | 2026-11-15 | independent | Source ↗ |
| Granite 4.2 3B | 0% | 2026-11-15 | independent | Source ↗ |
| Claude Fable 5.2 | 73.3% | 2026-11-15 | independent | Source ↗ |
| GPT-6.1 Terra | 57.6% | 2026-11-15 | independent | Source ↗ |
| Gemini 4 Carbon | 64.8% | 2026-11-15 | independent | Source ↗ |
| Inkling | 0% | 2026-11-15 | independent | Source ↗ |
| Granite 4.2 30B | 0% | 2026-11-15 | independent | Source ↗ |
| K2 Horizon 7B | 0% | 2026-11-15 | independent | Source ↗ |
| MiMo-V2.6-Flash | 4.3% | 2026-11-15 | independent | Source ↗ |
| Mistral Small 4 | 0% | 2026-11-15 | independent | Source ↗ |
| Solar Pro 4 | 0% | 2026-11-15 | independent | Source ↗ |
| K-EXAONE 2.0 | 0% | 2026-11-15 | independent | Source ↗ |
| Claude Sonnet 5.5 | 53.3% | 2026-11-15 | independent | Source ↗ |
| GPT-6 Luna | 8.6% | 2026-11-15 | independent | Source ↗ |
| North Mini Code | 0% | 2026-11-15 | independent | Source ↗ |
| GPT-6 Astra | 63.3% | 2026-11-15 | independent | Source ↗ |
| Claude Haiku 5.5 | 20% | 2026-11-15 | independent | Source ↗ |
| Grok 4.7 | 11.4% | 2026-11-15 | independent | Source ↗ |
| MiMo-V2.6-Pro | 5.7% | 2026-11-15 | independent | Source ↗ |
| MiniMax M3 | 0.5% | 2026-11-15 | independent | Source ↗ |
| Mistral Large 4 | 4.3% | 2026-11-15 | independent | Source ↗ |
| Nemotron 3.5 Lightning | 0% | 2026-11-15 | independent | Source ↗ |
| Command A+ | 0% | 2026-11-15 | independent | Source ↗ |
| Claude Opus 5.5 | 59% | 2026-11-15 | independent | Source ↗ |
| K2 Horizon 3.7B | 0% | 2026-11-15 | independent | Source ↗ |
| Qwen3.8 27B | 0.5% | 2026-11-15 | independent | Source ↗ |
| Muse Spark Open | 1.9% | 2026-11-15 | independent | Source ↗ |
Human Baseline & Difficulty Horizon
Calibrated human reference points, specialist benchmarks, and ceiling thresholds.No peer-reviewed human baseline has been published for this benchmark.
Metric & Scoring Methodology
Verification protocols, aggregation formulas, and specialized metric variants.task success (%)Dataset & Compute Cost
Evaluation volume, public availability, API pricing, and local hardware requirements.$5 – $20 USD for full benchmark evaluation run on frontier APIs.
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.lm_eval --model hf --model_args pretrained=<model_path> --tasks terminal-bench-science --batch_size autoopencompass --datasets terminal-bench-science --models <model_config># 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,
)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.Static fixed snapshot
Public on web / HuggingFace