Releases & self-play Elo
Self-play Elo
Estimated relative strength changes from internal self-play testing. These are not absolute engine ratings or Lichess ratings.
| Version | What changed | Compared with | Approx. Elo change |
|---|---|---|---|
| 1.1.10 | Removed an unnecessary evaluation term. | 1.1.9 | ≈ +5 |
| 1.1.9 | Improved endgame play and draw avoidance. | 1.1.8.2 | ≈ +33 |
| 1.1.8.2 | Fixed bishop evaluation in basic endgames. | 1.1.8.1 | ≈ +6 |
| 1.1.8.1 | Fixed opening evaluation and removed an unnecessary term. | 1.1.8 | ≈ -7 |
| 1.1.8 | Improved opening evaluation. | 1.1.1 | ≈ -9 |
| 1.1.7 | Added evaluation of development and king safety. | 1.1.6 | ≈ +2 |
| 1.1.6 | Recycled old search branches when memory fills. | 1.1.5 | ≈ -3 |
| Simulated memory pressure: recycling on vs off | ≈ +168 | ||
| 1.1.5 | Removed the legacy search code. | 1.1.4.1 | ≈ +0 |
| 1.1.4.1 | Fixed UCI command handling. | 1.1.4 | ≈ -0 |
| 1.1.4 | Improved king play in pawn endgames. | 1.1.3 | ≈ -0 |
| 1.1.3 | Improved rook-versus-pawn endgames. | 1.1.2 | ≈ +3 |
| 1.1.2 | Improved queen-versus-pawn endgames. | 1.1.1 | ≈ +6 |
| 1.1.1 | Improved bishop-and-knight checkmates. | 1.1.0.1 | ≈ +1 |
| 1.1.0.1 | Fixed illegal king captures in exchange evaluation. | 1.1.0 | ≈ +4 |
| Improved move selection and endgame play, while reusing analysis between moves. | 1.0.0 | ≈ +44Estimated summed gain | |
| 1.0.3 | Refined move exploration. | 1.0.2 | ≈ +11 |
| 1.0.2 | Improved conversion of winning endgames. | 1.0.1 | ≈ +10 |
| 1.0.1 | Improved exploration of candidate moves. | 1.0.0 | ≈ +23 |
| First complete engine: evaluates positions, searches for moves, and manages its clock. | — | Reference |