OpenAI GPT-6.2 Sol at launch
- OpenAI
- Released November 17, 2026
- 1.05M context
OpenAI released GPT-6.2 Sol on November 17, 2026. As of November 22, 2026, it ranks #14 of 120 models on the LLMBenchmark Index, between Anthropic Claude Opus 5 (#13) and OpenAI GPT-5.6 Sol (#15). OpenAI's own announcement reports 19 benchmark results, including GPQA Diamond (96.0%) and SEC-Bench Pro (88.3%). It costs $2.00 per million input tokens and $10.00 per million output tokens on OpenRouter.
What OpenAI reported at launch
From OpenAI's announcement, November 17, 2026. The maker's own numbers, not re-run by us.
-
AAV
0.6
Mean Spearman rank correlation (higher is better)
#3 of 9 models with a reported result
-
Coronavirus--ACE2
0.4
Median composite score (higher is better)
#3 of 9 models with a reported result
-
ExploitBench Internal Port (June--August 2026)
50.5%
Mean arbitrary code execution rate across V8 and Chrome
#2 of 5 models with a reported result
-
ExploitBench V8
100.0%
Mean capability credit (five attempts per vulnerability)
#1 of 5 models with a reported result
-
GPQA Diamond
96.0%
no tools; accuracy
#2 of 56 models with a reported result
-
Initial Stage
74.7%
mean rubric score
-
Initial Stage
74.7%
mean rubric score
#4 of 5 models with a reported result
-
MentalHealthBench (emergent)
52.6
#6 of 10 models with a reported result
-
MentalHealthBench (high acuity)
52.9
high-acuity
#7 of 10 models with a reported result
-
MentalHealthBench (non-acute)
49.4
#6 of 10 models with a reported result
-
MentalHealthBench (overall)
50.9
#7 of 10 models with a reported result
-
Multi-select Virology Troubleshooting
55.4%
pass@1; observed
Show all 19
-
Multi-select Virology Troubleshooting
55.7%
pass@1; refusal-adjusted
#3 of 6 models with a reported result
-
Phage--plasmid Co-evolution
12.4
Median of pair-weighted negative log-likelihood (lower is better)
#4 of 9 models with a reported result
-
SEC-Bench Pro
88.3%
Vulnerability reproduction (pass@1)
#1 of 15 models with a reported result
-
Secondary Stage
91.4%
mean reference percentile
#3 of 5 models with a reported result
-
Secondary Stage
14.6%
mean share of baseline training time saved
-
Secondary Stage
44.9%
mean share of performance gap closed
-
SHP2 protein function
0.4
Mean of median R² across three assays (higher is better)
#3 of 5 models with a reported result
Independent results so far
56 scores from 56 boards that test models themselves. Ranks compare GPT-6.2 Sol with every model on the same board.
General (14)
| Benchmark | Score | Rank | Setting | Source |
|---|---|---|---|---|
| Vals Index | 66.1% | #9 of 40 | — | Vals Index |
| LiveBench data analysis | 81.1% | #11 of 35 | — | LiveBench |
| MultiChallenge | 72.0% | #11 of 28 | high | Scale SEAL leaderboards |
| LMArena Text: instruction following | 1434 | #56 of 153 | Max | LMArena (Hugging Face dataset) |
| LMArena Text: writing and language | 1420 | #58 of 153 | Max | LMArena (Hugging Face dataset) |
| LMArena Text: entertainment and media | 1408 | #59 of 153 | Max | LMArena (Hugging Face dataset) |
| LMArena Text: creative writing | 1403 | #65 of 153 | Max | LMArena (Hugging Face dataset) |
| LMArena Text: longer queries | 1427 | #73 of 153 | Max | LMArena (Hugging Face dataset) |
Show all 14
| LMArena Text: multi-turn | 1420 | #80 of 153 | Max | LMArena (Hugging Face dataset) |
|---|---|---|---|---|
| LMArena Text: non-English | 1405 | #83 of 153 | Max | LMArena (Hugging Face dataset) |
| LMArena Text | 1414 | #88 of 153 | Max | LMArena (Hugging Face dataset) |
| LMArena Text: English | 1419 | #89 of 153 | Max | LMArena (Hugging Face dataset) |
| LMArena Text: Russian | 1401 | #89 of 153 | Max | LMArena (Hugging Face dataset) |
| LMArena Text: Chinese | 1434 | #94 of 147 | Max | LMArena (Hugging Face dataset) |
Agentic work (11)
| Benchmark | Score | Rank | Setting | Source |
|---|---|---|---|---|
| Terminal-Bench-Science (Vals AI) | 58.6% | #4 of 45 | Mini-SWE-agent | Vals AI boards |
| LiveBench research | 90.1% | #6 of 35 | — | LiveBench |
| Vals Science Index | 57.8% | #8 of 27 | — | Vals AI boards |
| BioMysteryBench (Vals AI) | 77.8% | #9 of 33 | — | Vals AI boards |
| LiveBench browser | 88.2% | #9 of 35 | — | LiveBench |
| LiveBench long agentic | 63.6% | #9 of 35 | — | LiveBench |
| APEX-Agents | 66.6% | #14 of 63 | Max | Epoch AI benchmarking hub |
| Harvey's Legal Agent Benchmark (Vals AI) | 12.1% | #20 of 40 | — | Vals AI boards |
Show all 11
| LMArena Search | 1179 | #20 of 29 | Max | LMArena (Hugging Face dataset) |
|---|---|---|---|---|
| Tax Agent Bench (Vals AI) | 24.8% | #47 of 83 | — | Vals AI boards |
| Legal Research Bench (Vals AI) | 37.0% | #49 of 89 | — | Vals AI boards |
Coding (11)
| Benchmark | Score | Rank | Setting | Source |
|---|---|---|---|---|
| LMArena WebDev: HTML | 1840 | #2 of 108 | Max | LMArena (Hugging Face dataset) |
| FrontierCode 2.0 | 44.4% | #5 of 62 | Max | Artificial Analysis |
| Terminal-Bench 4.0 (Artificial Analysis run) | 60.1% | #7 of 117 | Max | Artificial Analysis |
| LMArena WebDev | 1754 | #8 of 109 | Max | LMArena WebDev |
| Terminal-Bench 4 (Vals AI) | 60.6% | #8 of 63 | Mini-SWE-agent | Vals AI boards |
| GameBench 2 | 94.4 | #10 of 105 | xHigh | DuelLab GameBench 2 |
| LMArena WebDev: React | 1761 | #10 of 101 | Max | LMArena (Hugging Face dataset) |
| LiveBench coding | 83.7% | #11 of 35 | — | LiveBench |
Show all 11
| Vibe Code Bench (Vals AI) | 87.6% | #17 of 110 | OpenHands | Vals AI boards |
|---|---|---|---|---|
| LMArena Text: coding | 1440 | #85 of 153 | Max | LMArena (Hugging Face dataset) |
| LMArena Text: software and IT | 1434 | #91 of 153 | Max | LMArena (Hugging Face dataset) |
Reasoning (11)
| Benchmark | Score | Rank | Setting | Source |
|---|---|---|---|---|
| Humanity's Last Exam (Scale run) | 56.5% | #4 of 25 | high | Scale SEAL leaderboards |
| HLE-Diamond (Artificial Analysis run) | 64.0% | #6 of 79 | Max | Artificial Analysis |
| CritPt v2, hard subset (Artificial Analysis run) | 64.0% | #10 of 62 | Medium | Artificial Analysis |
| ARC-AGI | 97.5% | #11 of 77 | Max | Epoch AI benchmarking hub |
| MysteryMechanism (Vals AI) | 36.9% | #11 of 32 | — | Vals AI boards |
| LiveBench reasoning | 90.6% | #12 of 35 | — | LiveBench |
| ARC-AGI-2 | 85.8% | #14 of 78 | xHigh | Epoch AI benchmarking hub |
| Kagi LLM Benchmark | 67.0% | #19 of 65 | — | Kagi LLM Benchmarking Project |
Show all 11
| SimpleBench | 69.9% | #29 of 78 | — | Epoch AI benchmarking hub |
|---|---|---|---|---|
| LMArena Text: hard prompts | 1419 | #92 of 153 | Max | LMArena (Hugging Face dataset) |
| LMArena Text: hard prompts in English | 1413 | #99 of 153 | Max | LMArena (Hugging Face dataset) |
Web design (4)
| Benchmark | Score | Rank | Setting | Source |
|---|---|---|---|---|
| Design Arena: all code categories | 1405 | #1 of 134 | — | Design Arena |
| Design Arena: 3D | 1482 | #4 of 127 | — | Design Arena |
| Design Arena: game dev | 1437 | #4 of 134 | — | Design Arena |
| LiveBench design | 80.0% | #8 of 35 | — | LiveBench |
Knowledge (3)
| Benchmark | Score | Rank | Setting | Source |
|---|---|---|---|---|
| AA-Omniscience | 40.4 | #11 of 65 | Max | Artificial Analysis |
| LMArena Text: science | 1432 | #81 of 153 | Max | LMArena (Hugging Face dataset) |
| LMArena Text: business and finance | 1383 | #103 of 153 | Max | LMArena (Hugging Face dataset) |
Intelligence (1)
| Benchmark | Score | Rank | Setting | Source |
|---|---|---|---|---|
| AA Intelligence Index | 57.3 | #6 of 65 | — | Artificial Analysis (via OpenRouter) |