CHESS BENCHMARK · LOCAL, NO THINKING

Gemma 3 4B

Google · open weights · released 2025-03-12 · route gemma3:4b · Model profile

Opening gap: +206 (95%: -641 to +1,053). The interval includes zero, so we can't yet say whether it performs differently across the variants.

Standard

Rating
71395%: 82 to 1,343
Record
0 won, 0 drawn, 2 lost of 2
Accuracy (open positions)
76.4
Accuracy (all moves)
76.4
Average centipawn loss
89
Blunders / mistakes / inaccuracies per 100 moves
28.6 / 0.0 / 0.0
Matched Stockfish's first choice
14%
In Stockfish's top three
29%
Legal on the first try
66.7%
Illegal / invalid
8 of 15 answers
Forfeits
2
Adjudicated games
0
API-equivalent cost per game
n/a
API-equivalent cost per move
n/a
Seconds per move
15.2
Model minutes per game
1.1
Tokens per move (in / out)
354 / 16
Reasoning tokens per move
n/a
Aborted on our side (not counted)
0

By phase

PhaseMovesAccuracyCentipawn lossBlunders /100Engine match
Opening776.48928.614%
Middlegame0n/an/an/an/a
Endgame0n/an/an/an/a

Chess960

Rating
50695%: -59 to 1,072
Record
0 won, 0 drawn, 2 lost of 2
Accuracy (open positions)
83.3
Accuracy (all moves)
83.3
Average centipawn loss
47
Blunders / mistakes / inaccuracies per 100 moves
0.0 / 0.0 / 50.0
Matched Stockfish's first choice
25%
In Stockfish's top three
25%
Legal on the first try
40.0%
Illegal / invalid
12 of 20 answers
Forfeits
2
Adjudicated games
0
API-equivalent cost per game
n/a
API-equivalent cost per move
n/a
Seconds per move
20.5
Model minutes per game
1.7
Tokens per move (in / out)
560 / 24
Reasoning tokens per move
n/a
Aborted on our side (not counted)
0

By phase

PhaseMovesAccuracyCentipawn lossBlunders /100Engine match
Opening883.3470.025%
Middlegame0n/an/an/an/a
Endgame0n/an/an/an/a

Games