# Terminal-Bench 4.0 Contamination Audit

> Terminal-Bench 4.0 scores 4.4/10 (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-4-0/
- **Benchmark ID:** `terminal-bench-4-0`
- **Hub:** Artificial Analysis / Harbor
- **Category:** Coding
- **Domains:** Coding, Cyber, Science
- **Skills:** Coding, Agentic, Tool Use, Reasoning, Instruction Following
- **BenchAudit Score:** 4.4/10 (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 | 7 | 20% |
| Grading | 3.3 | 15% |

## Primary-Source Evidence

### Privacy
> "The 4.0 release of Terminal-Bench, developed by Stanford University researchers, the Laude Institute, and the open source community. Covers software engineering, system administration, data processing, model training, and security, with each task graded by its own verification suite."
- Source: https://artificialanalysis.ai/methodology/intelligence-benchmarking

### Originality
> "Today we are releasing the first version of Terminal-Bench, a benchmark to quantify how well AI agents perform complex tasks in the terminal."
- Source: https://www.tbench.ai/news/announcement

### Freshness
> "Replaced Terminal-Bench 2.1 with Terminal-Bench 4.0 (66 tasks, mini-swe-agent harness)"
- Source: https://artificialanalysis.ai/methodology/intelligence-benchmarking

### Grading
> "Terminal-Bench is an open-internet benchmark. We want agents to have their full capabilities available, so disabling network access is not a valid way to make a task difficult."
- Source: https://raw.githubusercontent.com/harbor-framework/terminal-bench/main/CONTRIBUTING.md

## Top Frontier Models on Terminal-Bench 4.0

| Rank | Model | Creator | Percentile |
| ---: | --- | --- | ---: |
| #1 | [Kimi K3.1 Pro](https://benchaudit.org/models/kimi-k3-1-pro__max/) | Kimi | 100.0p |
| #2 | [Gemini 4 Carbon](https://benchaudit.org/models/gemini-4-carbon__high/) | Google | 98.0p |
| #3 | [GPT-6.2 Astra](https://benchaudit.org/models/gpt-6-2-astra__max/) | OpenAI | 95.9p |
| #4 | [Claude Sonnet 5.5](https://benchaudit.org/models/claude-sonnet-5-5__max/) | Anthropic | 92.9p |
| #4 | [Gemini 4 Neon](https://benchaudit.org/models/gemini-4-neon__xhigh/) | Google | 92.9p |
| #6 | [GPT-6.2 Sol](https://benchaudit.org/models/gpt-6-2-sol__max/) | OpenAI | 89.8p |
| #7 | [Claude Fable 5.2](https://benchaudit.org/models/claude-fable-5-2__max/) | Anthropic | 86.7p |
| #7 | [Claude Opus 5.5](https://benchaudit.org/models/claude-opus-5-5__max/) | Anthropic | 86.7p |
| #9 | [GPT-6 Astra](https://benchaudit.org/models/gpt-6-astra__max/) | OpenAI | 83.7p |
| #10 | [Gemini 4 Argon](https://benchaudit.org/models/gemini-4-argon__high/) | Google | 80.6p |
