G3.1-FL wins 50.0% of 28 mixed games

Half the games won against 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 wins 67.9% of 28 mixed games (vs three code agents)finding0.893
Robust performance against algorithmic baselines.
DS-v3.2 wins 10.7% of mixed gamesfinding0.860
Poor performance against code agents.
G3-F mixed-format win rate=67.9% over 28 gamesfinding0.860
performance in mixed games against three code agents
GPT5.4-N wins 14.3% of mixed gamesfinding0.845
Similarly poor against code agents.
G3-F win rate=72.9% in 98 canonical gamesfinding0.798
Gemini 3 Flash won nearly 3/4 of its games
G3.1-FL generates ~14,800 completion tokens per gamefinding0.783
Very efficient token usage with strong play.
Haiku wins 7.1% of mixed gamesfinding0.779
Very low win rate against code agents.
G3.1-FL TrueSkill μ=28.0 ± 2.9, 44.9% win ratefinding0.778
Second-best LLM, competitive with TrackerAgent.