Qwen3-32B

Weak-tier open-source model exhibiting both harness activation failure and adherence failure, with 25.1% skill-load rate

Neighborhood — ranked by edge-count

paper

concept

Qwen3-1.7B
related_to
Smallest Qwen3 model tested; used in conscientiousness sweep example (Table 6)
Qwen3-4B
related_to
4B Qwen3 model tested in OCEAN benchmarks
Qwen3-235B-A22B
related_to
Large open-source model used as anchor agent and anchor evolver; illustrates benchmark-dependent evolver performance

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.

Qwen3-14Bconcept0.846
14B Qwen3 model quantized to 4-bit NF4; tested in OCEAN benchmarks
Qwen3.5-9Bconcept0.840
Smallest model tested as evolver; produces harness updates comparable to Claude Opus 4.6 on SkillsBench
Qwen 3 0.6B Embeddingmethod0.792
Embedding model used to embed user messages for ridge regression analysis of persona drift causes
Qwen2.5-VL-7Bconcept0.789
Base vision-language model used to instantiate ATLAS.
Qwen 3 32B is most likely to hallucinate human personas (names, birthplaces, years of experience) when steered away from the Assistantfinding0.713
Model-specific difference in how steered personas manifest
Qwen 35B (3B active params, score 4.38) outscores Hermes 405B (405B active params, score 1.75) by 2.5xfinding0.705
Parameters don't predict scores; 135x more parameters yields 60% lower score
Optimal activation capping layers for Qwen 3 32B are layers 46-53 (out of 64) at 25th percentile capfinding0.703
Specific implementation finding for Qwen capping parameters
Qwen3-32B adherence drops from 0.52 after harness loading to 0.13 at final validation (drift of -0.39)finding0.701
Demonstrates long-horizon instruction-following bottleneck for weak-tier models