Haiku overbid rate=0.87%

second highest overbid rate

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.

Claude Haiku 4.5 overbid rate 0.87%finding0.882
Haiku's overbid frequency is second highest after G2.5-FL.
G3-F overbid rate=0.00%finding0.777
Gemini 3 Flash never overbid
Haiku wins 7.1% of mixed gamesfinding0.774
Very low win rate against code agents.
GPT5.4-N overbid rate=0.47%finding0.770
overbid rate for GPT-5.4 Nano
G2.5-FL overbid rate=1.20%, highest among all agentsfinding0.769
highest overbid frequency observed
DS-v3.2 overbid rate=0.49%finding0.755
overbid rate for DeepSeek v3.2
Haiku TrueSkill μ=21.8±2.8finding0.727
Claude Haiku 4.5 TrueSkill rating
Haiku test-retest score delta is 0.02 (6.47 vs 6.49) across two full 30-koan battery runsfinding0.723
Demonstrates high stability for Anthropic API models