G3-F mixed-format win rate=67.9% over 28 games

performance in mixed games against three code agents

Source paper

extracted_from

(2026) · Robert Müller · Clemens Müller

cosine ≥ 0.65 · no typed edge

Entities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.

G3-F win rate=72.9% in 98 canonical gamesfinding0.894
Gemini 3 Flash won nearly 3/4 of its games
G3-F wins 67.9% of 28 mixed games (vs three code agents)finding0.861
Robust performance against algorithmic baselines.
G3.1-FL wins 50.0% of 28 mixed gamesfinding0.860
Half the games won against code agents.
DS-v3.2 wins 10.7% of mixed gamesfinding0.806
Poor performance against code agents.
GPT5.4-N wins 14.3% of mixed gamesfinding0.802
Similarly poor against code agents.
G3.1-FL TrueSkill μ=28.0 ± 2.9, 44.9% win ratefinding0.795
Second-best LLM, competitive with TrackerAgent.
G3-F overbid rate=0.00%finding0.791
Gemini 3 Flash never overbid
G3-F TrueSkill μ=30.1 ± 3.3, 72.9% wins, median score 5,250 on combined-comp1 slice (n=98 canonical games)finding0.786
Top LLM performance with high win rate and large score.