SEAL Showdown: leaderboard
Scale AI human expert evaluation of LLMs across real-world tasks. Arena-style scoring based on pairwise comparisons by trained human raters.
Metric: Arena Score. Source: scale.com. Saturation forecast: Not forecast. 47 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Pro (Preview) | 1255.4 | #66 |
| 2 | Gemini 3 Flash | 1206.5 | #99 |
| 3 | Voxtral-Small-24B-2507 | 1173.7 | #581 |
| 4 | GPT-5.5 (2026-04-23) | 1116.4 | #22 |
| 5 | Claude Opus 4.7 (Thinking) | 1113.1 | #49 (Claude Opus 4.7) |
| 6 | Claude Opus 4.6 (Thinking) | 1109 | #63 (Claude Opus 4.6) |
| 7 | GPT-5.2 Instant | 1097.9 | #210 |
| 8 | Claude Opus 4.1 (20250805) (Thinking) | 1093 | #132 (Claude Opus 4.1 (20250805)) |
| 9 | Claude Opus 4.7 | 1087.1 | #49 |
| 10 | Claude Opus 4.6 | 1086.1 | #63 |
| 11 | Kimi K2.5 (Thinking) | 1085.1 | #144 (Kimi K2.5) |
| 12 | GPT-5 Chat | 1081.5 | #241 |
| 13 | DeepSeek V3.2 | 1077.7 | #205 |
| 14 | Claude Sonnet 4.6 | 1072.4 | #90 |
| 15 | Claude Opus 4.5 (20251101) | 1066 | #74 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=seal-showdown · How It Works · Data refreshed daily, snapshot 2026-10-20.