# Terminal-Bench-Science 0.1 Contamination Audit

> Terminal-Bench-Science 0.1 scores 4.2/10 (Tier C, High Risk) on BenchAudit as of November 2026, reflecting publicly downloadable plaintext test files and self-reported or live-network evaluation.

- **Canonical URL:** https://benchaudit.org/benchmarks/terminal-bench-science-0-1/
- **Benchmark ID:** `terminal-bench-science-0-1`
- **Hub:** Artificial Analysis / Stanford / Harbor
- **Category:** Science
- **Domains:** Science, Coding
- **Skills:** Coding, Agentic, Tool Use, Reasoning, Instruction Following
- **BenchAudit Score:** 4.2/10 (Tier C - High Risk)
- **Last Audited:** 2026-11-17 (Ref 2026-11)

## Factor Scores

| Factor | Score (/10) | Weight |
| --- | ---: | ---: |
| Privacy | 0 | 40% |
| Originality | 10 | 25% |
| Freshness | 6 | 20% |
| Grading | 3.3 | 15% |

## Primary-Source Evidence

### Privacy
> "Dataset: https://github.com/harbor-framework/terminal-bench-science Implementation: We evaluate the full Terminal-Bench-Science 0.1.0 release (70 tasks: 19 life sciences, 17 physical sciences, 17 mathematical sciences, 9 engineering sciences, 8 earth sciences) using the mini-swe-agent harness, with pass@1 scoring averaged over 3 repeats per task"
- Source: https://artificialanalysis.ai/methodology/intelligence-benchmarking

### Originality
> "Terminal-Bench-Science is an open academic collaboration developed by Stanford University researchers with the Terminal-Bench and Harbor team. It includes contributions from scientists at institutions across the globe, comprising 70 expert-curated and reviewed research workflow tasks drawn from real scientific practice."
- Source: https://artificialanalysis.ai/evaluations/terminal-bench-science

### Freshness
> "Written by Steven Dillmann on August 27, 2026 Terminal-Bench-Science is a benchmark led by researchers at Stanford University and built by the team behind Terminal-Bench in collaboration with domain experts from a range of scientific disciplines and research institutions around the world."
- Source: https://www.terminal-bench-science.ai/announcement

### Grading
> "network_mode = "public" # anti-cheat rests on the scrubbed lake identity, not on network isolation"
- Source: https://raw.githubusercontent.com/harbor-framework/terminal-bench-science/v0.1.0/tasks/earth-sciences/geosciences/mendota-ice-phenology/task.toml

## Top Frontier Models on Terminal-Bench-Science 0.1

| Rank | Model | Creator | Percentile |
| ---: | --- | --- | ---: |
| #1 | [GPT-6.2 Astra](https://benchaudit.org/models/gpt-6-2-astra__max/) | OpenAI | 100.0p |
| #2 | [Claude Fable 5.2](https://benchaudit.org/models/claude-fable-5-2__max/) | Anthropic | 100.0p |
| #3 | [Gemini 4 Neon](https://benchaudit.org/models/gemini-4-neon__xhigh/) | Google | 100.0p |
| #4 | [GPT-6.2 Sol](https://benchaudit.org/models/gpt-6-2-sol__max/) | OpenAI | 97.5p |
| #5 | [Gemini 4 Carbon](https://benchaudit.org/models/gemini-4-carbon__high/) | Google | 95.0p |
| #6 | [GPT-6 Astra](https://benchaudit.org/models/gpt-6-astra__max/) | OpenAI | 92.4p |
| #7 | [Claude Opus 5.5](https://benchaudit.org/models/claude-opus-5-5__max/) | Anthropic | 89.9p |
| #8 | [GPT-6.1 Sol](https://benchaudit.org/models/gpt-6-1-sol__max/) | OpenAI | 87.4p |
| #9 | [GPT-6.1 Terra](https://benchaudit.org/models/gpt-6-1-terra__max/) | OpenAI | 84.9p |
| #10 | [Claude Sonnet 5.5](https://benchaudit.org/models/claude-sonnet-5-5__max/) | Anthropic | 82.4p |
