AI Chess Leaderboard (Continuation): leaderboard
One of two modes on the dubesor AI chess project. The model gets only raw PGN movetext and must continue the game, a format the site notes draws illegal move attempts; games average about 48 turns at default model settings. Ratings start from Stockfish 17.1 accuracy over a model's first ten games, then update by standard Elo on games between different models, so they are relative to this pool; higher is better.
Metric: Elo. Source: dubesor.de. Saturation forecast: Around November 2027. 284 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Pro (Preview) | 1810 | #66 |
| 2 | GPT-4.5 (Preview) | 1800 | #244 |
| 3 | Gemini 3.8 Flash | 1747 | #21 |
| 4 | Gemini 3.7 Flash | 1746 | #23 |
| 5 | Gemini 3.1 Pro (Preview) | 1732 | #58 |
| 6 | Gemini 3 Flash (Preview) | 1732 | #82 |
| 7 | Gemini 3.6 Flash | 1732 | #51 |
| 8 | Gemini 3.5 Flash | 1706 | #60 |
| 9 | Gemini 4 Argon | 1641 | #8 |
| 10 | GPT-4o ChatGPT | 1584 | |
| 11 | GPT-6.1 Sol | 1530 | #7 |
| 12 | GPT-6 Astra | 1520 | #6 |
| 13 | GPT-4o | 1496 | #343 |
| 14 | Gemini 3.5 Flash Lite | 1476 | #156 |
| 15 | GPT-5 | 1471 | #98 |
No result here: #10 Muse Wonder Full.
Interactive version: theaggregate.ai/benchmark?slug=ai-chess-leaderboard-continuation · How It Works · Data refreshed daily, snapshot 2026-10-20.