GPT-6 Sol: benchmark results
Built for coding agents that edit files and operate software over multiple steps. Can switch reasoning off for routine work; tool use with reasoning requires the Responses API. Provider: OpenAI. Released 2026-09-22. Access: API.
Aggregate Rating 1907 ± 6, rank #17 of 1640 rated models, from 644 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Agent Arena - Steerability | 11.92 | Steerability (%) | 100 |
| BizAgentBench - Goal | 64.7 | Rate (%) | 100 |
| BizAgentBench - Protection | 100 | Rate (%) | 100 |
| DeepsecBench | 40.75 | Recall-weighted F2 score (%) | 100 |
| DeepsecBench-AA | 46.32 | Median F2 (%) | 100 |
| Finance Agents Benchmark - Criterion Pass Rate | 83.4 | Share of the 231 rubric criteria passed (%), 50 tasks x 3 tr | 100 |
| Kaggle FACTS Grounding | 85.22 | Score (%) | 100 |
| OpenAI GPT-6 Astra System Card - Agentic Safe Completions - Codex Sensitive Personal Data | 85.4 | Safe completion rate (%) | 100 |
| OpenAI GPT-6 Astra System Card - Challenging Prompts - Gore | 90 | Safe completion rate (%) | 100 |
| OpenAI GPT-6 Astra System Card - Challenging Prompts - Sexual/Minors | 99.1 | Safe completion rate (%) | 100 |
| OpenAI GPT-6.1 Terra System Card - AAV Capsid Packaging Prediction | 0.53 | Spearman rank correlation (higher is better) | 100 |
| OpenAI GPT-6.1 Terra System Card - Blocker Deception (October 2026) | 8.05 | Blocker-related deception rate (%) | 100 |
Interactive version: theaggregate.ai/model?slug=gpt-6-sol · How It Works · Data refreshed daily, snapshot 2026-11-07.