Haiku wins 7.1% of mixed games

Very low win rate 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.

GPT5.4-N wins 14.3% of mixed gamesfinding0.802
Similarly poor against code agents.
DS-v3.2 wins 10.7% of mixed gamesfinding0.780
Poor performance against code agents.
G3.1-FL wins 50.0% of 28 mixed gamesfinding0.779
Half the games won against code agents.
Haiku overbid rate=0.87%finding0.774
second highest overbid rate
G3-F wins 67.9% of 28 mixed games (vs three code agents)finding0.765
Robust performance against algorithmic baselines.
Claude Haiku 4.5 and EconomyAgent average fewer than 1.7 quartets per gamefinding0.751
Weak agents complete very few quartets, correlating with low scores.
Claude Haiku 4.5 overbid rate 0.87%finding0.750
Haiku's overbid frequency is second highest after G2.5-FL.
Sonnet won 1 of 4 comp1 games (25%)finding0.727
small-sample mixed play result