finding

active

finding:claude-haiku-4-5-and-economyagent-average-fewer-than-1-7-quartets-per-game

Claude Haiku 4.5 and EconomyAgent average fewer than 1.7 quartets per game

Weak agents complete very few quartets, correlating with low scores.

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 averages 3.96 quartets per gamefinding0.793
high quartet completion rate
Claude Haiku 4.5 overbid rate 0.87%finding0.784
Haiku's overbid frequency is second highest after G2.5-FL.
Haiku wins 7.1% of mixed gamesfinding0.751
Very low win rate against code agents.
On SWE-bench, Claude Opus 4.6 and Claude Sonnet 4.6 both achieve 7.4 pp harness-updating gain; Claude Haiku 4.5 achieves 8.0 ppfinding0.745
Full evolver-side SWE results showing comparable performance across Claude family tiers
G2.5-FL cost per quartet 1,193 coinsfinding0.738
Much higher cost per quartet due to waste.
Gemini 3 Flash averages 3.96 quartets per gamefinding0.735
G3-F completed on average 3.96 quartets per game.
Claude Haiku 4.5 and GPT-5.4 Nano have TC tightness τ ≈ 0.4, the tightest among allfinding0.726
These two LLMs bargain with minimal overpayment but low overall efficiency.
Claude 3.5 Sonnet shows higher rate of alignment-faking reasoning than Claude 3 Opus in helpful-only setting but almost none in animal welfarefinding0.726
Linked to Claude 3.5 Sonnet not exhibiting pro-animal-welfare preferences