GPT-6.2 Sol vs Claude Opus 5.5: Opus Still Has the Crown
OpenAI has officially announced its latest model, GPT-6.2 Sol, to compete with Claude Opus 5.5. In overall intelligence, Opus 5.5 still leads and narrowly wins, scoring 59.1 versus 57.3 on the Artificial Analysis Intelligence Index. However, OpenAI once again bets on efficiency and price, and GPT-6.2 Sol beats Claude Opus 5.5 by a huge margin there, at about 90% cheaper per Artificial Analysis task. Here are the detailed benchmarks to help you see the real differences between these models.
OpenAI's benchmarks at max effort | |||
| Benchmark | Difference | ||
|---|---|---|---|
| Agents' Last Exam v1.1 | 73.0% | 71.6% | |
| Terminal-Bench-Science 0.1 | 68.0% | 63.3% | |
| FrontierCode 2.0 | 43.8% | 45.5% | |
| GDPval v2 (OpenAI internal) | 46.9% | 43.5% | |
| Harvey LAB-AA v1.1 | 11.1% | 4.2% | |
| BenchCAD 2.1 Core | 73.1% | 65.3% | |
| AutomationBench 1.0.6 | 41.2% at high | 42.5% | |
| Vals Science Index v1 | 58.5% at xhigh | 58.7% | |
Artificial Analysis benchmarks at max effort | |||
| Intelligence Index v5.0.1 | 57.3 | 59.1 | |
| AA-Briefcase v1.1 | 1612 Elo | 1800 Elo | |
| AutomationBench-AA | 71.3% | 69.5% | |
| Terminal-Bench 4.0 | 60.1% | 59.6% | |
| FrontierCode-AA 2.0 | 44.4% | 46.4% | |
| AA-Omniscience v1.1 | 40.4 | 48.1 | |
| GDP.pdf | 30.0% | 26.2% | |
| HLE-Diamond | 64.0% | 67.9% | |
| CritPt v2 (Hard Subset) | 62.5% | 68.0% | |
| Terminal-Bench-Science 0.1 | 67.6% | 59.0% | |
Intelligence Index by reasoning effort | |||
| Max | 57 | 59 | |
| Xhigh | 56 | 57 | |
| High | 55 | 52 | |
| Medium | 53 | 48 | |
| Low | 41 | 35 | |
Cost per Intelligence Index task by reasoning effort | |||
| Max | $1.01 | $10.04 | |
| Xhigh | $0.65 | $6.05 | |
| High | $0.56 | $3.46 | |
| Medium | $0.41 | $2.45 | |
| Low | $0.18 | $0.98 | |
Speed and tokens at max effort | |||
| Output tokens per task | 40.4K | 173.3K | |
| Reasoning tokens per task | 22.8K | 120.5K | |
| Time per task | 7.9 min | 24.5 min | |
| Output speed | 85 tokens/s | 118 tokens/s | |
| Time to first answer token | 62 s | 355 s | |
Pricing and specs | |||
| Input, per 1M tokens | $2.00 | $4.00 | |
| Output, per 1M tokens | $10.00 | $20.00 | |
| Cached input, per 1M tokens | $0.10 | $0.20 | |
| Cache writes, per 1M tokens | $2.50 for 30 min | $5.00 for 5 min | |
| Batch input, per 1M tokens | $1.00 | $2.00 | |
| Batch output, per 1M tokens | $5.00 | $10.00 | |
| Input over 272K tokens, per 1M | $4.00 | $4.00 | |
| Output over 272K tokens, per 1M | $15.00 | $20.00 | |
| Context window | 1.05M tokens | 1M tokens | |
| Max input | 922K tokens | Not listed | |
| Max output | 128K tokens | 128K tokens | |
| Max output in batch | 128K tokens | 300K tokens in beta | |
| Reasoning levels | Low to max | Low to max with adaptive thinking | |
| Text and image input | Yes | Yes | |
| Audio and video input | No | No | |
| Knowledge cutoff | June 1, 2026 | June 2026 | |
| Released | November 17, 2026 | September 22, 2026 | |