GPT-6.2 Sol vs GPT-6.1 Sol: Did OpenAI Deliver?
OpenAI has officially launched its latest flagship models, GPT-6.2 Astra and GPT-6.2 Sol. It is OpenAI's most muted release since GPT-5.6, and while GPT-6.2 Sol is not a big jump over its predecessor, it delivers better intelligence more efficiently and scores higher on several benchmarks. Sol is used by many, so here is how it compares with the previous generation, GPT-6.1 Sol.
OpenAI's benchmarks at max effort | |||
| Benchmark | Difference | ||
|---|---|---|---|
| Agents' Last Exam v1.1 | 73.0% | 72.5% | 6.2 Sol+0.5 pts |
| GDPval v2 (OpenAI internal) | 46.9% | 41.5% | 6.2 Sol+5.4 pts |
| AutomationBench 1.0.6 | 41.2% at high | 36.1% | 6.2 Sol+5.1 pts |
| LifeVal | 83.8% | 79.5% | 6.2 Sol+4.3 pts |
| Terminal-Bench-Science 0.1 | 68.0% | 57.0% | 6.2 Sol+11.0 pts |
| Vals Science Index v1 | 58.5% at xhigh | 59.6% | 6.1 Sol+1.1 pts |
| FrontierCode 2.0 | 43.8% | 39.0% | 6.2 Sol+4.8 pts |
| SecureSWE (OpenAI internal) | 55.6% | 49.7% | 6.2 Sol+5.9 pts |
| PatchBench | 75.6% | 71.0% | 6.2 Sol+4.6 pts |
| BenchCAD 2.1 Core | 73.1% | 52.6% | 6.2 Sol+20.5 pts |
Artificial Analysis benchmarks at max effort | |||
| Intelligence Index v5.0.1 | 57.3 | 54.6 | 6.2 Sol+2.7 |
| AA-Briefcase v1.1 | 1612 Elo | 1555 Elo | 6.2 Sol+57 Elo |
| AutomationBench-AA | 71.3% | 64.9% | 6.2 Sol+6.4 pts |
| Terminal-Bench 4.0 | 60.1% | 56.1% | 6.2 Sol+4.0 pts |
| FrontierCode-AA 2.0 | 44.4% | 40.6% | 6.2 Sol+3.8 pts |
| AA-Omniscience v1.1 | 40.4 | 42.0 | 6.1 Sol+1.6 |
| GDP.pdf | 30.0% | 31.0% | 6.1 Sol+1.0 pt |
| HLE-Diamond | 64.0% | 58.8% | 6.2 Sol+5.2 pts |
| CritPt v2 (Hard Subset) | 62.5% | 69.5% | 6.1 Sol+7.0 pts |
| Terminal-Bench-Science 0.1 | 67.6% | 58.1% | 6.2 Sol+9.5 pts |
Intelligence Index by reasoning effort | |||
| Max | 57 | 55 | 6.2 Sol+2 |
| Xhigh | 56 | 52 | 6.2 Sol+4 |
| High | 55 | 50 | 6.2 Sol+5 |
| Medium | 53 | 46 | 6.2 Sol+7 |
| Low | 41 | 35 | 6.2 Sol+6 |
Speed and cost at max effort | |||
| Cost per Intelligence Index task | $1.01 | $1.27 | 6.2 Sol20% cheaper |
| Time per task | 7.9 min | 23.3 min | 6.2 Sol66% faster |
| Output speed | 85 tokens/s | 49 tokens/s | 6.2 Sol73% faster |
| Time to first answer token | 62 s | 145 s | 6.2 Sol57% lower |
| Output tokens across the Index | 60.9M | 63.7M | 6.2 Sol4% fewer |
Pricing and specs | |||
| Input, per 1M tokens | $2.00 | $2.00 | |
| Cached input, per 1M tokens | $0.10 | $0.10 | |
| Cache writes, per 1M tokens | $2.50 | $2.50 | |
| Output, per 1M tokens | $10.00 | $10.00 | |
| Input over 272K tokens, per 1M | $4.00 | $4.00 | |
| Output over 272K tokens, per 1M | $15.00 | $15.00 | |
| Ultrafast until February 28, 2027 | 3x standard | 6x standard | 6.2 Sol50% cheaper |
| Ultrafast from March 1, 2027 | 6x standard | 6x standard | |
| Context window | 1.05M tokens | 1.05M tokens | |
| Max output | 128K tokens | 128K tokens | |
| Reasoning levels | Low to max | Low to max | |
| Text and image input | Yes | Yes | |
| Audio and video input | No | No | |
| Computer use and web search | Yes | Yes | |
| Fast mode | Yes | Yes | |
| Knowledge cutoff | June 1, 2026 | April 30, 2026 | 6.2 Sol32 days newer |
| Released | November 17, 2026 | September 29, 2026 | 6.2 Sol49 days newer |