finding
active
finding:gemini-3-flash-completes-fourth-quartet-by-paying-far-above-face-value-netting-1-800-points-from-multiplicative-scoringGemini 3 Flash completes fourth quartet by paying far above face value, netting ≈1,800 points from multiplicative scoring
A trace shows G3-F turning a nominally wasteful overpay into a net score gain due to the multiplicative formula.
Source paper
extracted_from(2026) · Robert Müller · Clemens Müller
Neighborhood — ranked by edge-count
Claims (1)
claim
- The benchmark’s diagnostic value lies in identifying why a model loses, not just that it losessupportsargues for fine-grained behavioral analysis over aggregate rankings
Related by similarity (8)
cosine ≥ 0.65 · no typed edgeEntities in the same semantic neighborhood but without a typed relation to this one — candidates for new edges or unrecognized duplicates.
- G3-F completed on average 3.96 quartets per game.
- G2.5-FL's cost per quartet is nearly double that of efficient agents.
- G3-F's median score across games was 5,250 points.
- In the 172-game exp2 slice, G3-F has the highest LLM win rate against deterministic baselines.
- The Bayesian skill posterior mean for G3-F is 30.1 with 3σ confidence interval ±3.3.
- G3-F achieved a win rate of 72.9% in the combined-comp1 98-game slice.
- Specific result for Gemini 2.0 Flash in Experiment 1; lowest rate among tested models
- G3-F escalates bidding by nearly 10× from early to late game, adapting to phase.