Qwen3.5-4B · OTIS Mock AIME 2024-2025 · Best score (across scorers) · % ↗
Qwen / MODEL EVIDENCE
Qwen3.5-4B
2 benchmarks · 2 observations · 2 sources · released 2026-02-24
Scores at a glance
2 benchmarks · large dot: Qwen3.5-4B; small dots: closest models where reported
Qwen3.5-4B (Thinking) · Scale Labs · small ↗
Closest models by score
Similar positions on shared benchmark leaderboards. Smaller gaps are closer.
Gemma 3 4B →Google
2 of 2 shared benchmark comparisons
How similarity is measured
Mean absolute gap in percentile positions on shared benchmark settings; lower is closer. At least 60% overlap is required, with a minimum of three when available. Independent evaluations are used when there are at least three; otherwise published report comparisons are used. Repeated model observations within one setting are averaged before ranking. Writing-style traits and composite indices are excluded. This describes similarity, not an overall quality ranking.
POST-TRAINING / SOURCE DISCLOSURE
What the Qwen3.5 post-trained family report says
The reviewed card describes progressively harder agent environments and asynchronous reinforcement-learning infrastructure. Its architecture and multimodal training are documented separately from its post-training objective.
- Agent environments with increasing task complexity
- Large-scale asynchronous reinforcement learning
The card reviewed is for 397B-A17B. It does not name a complete policy objective; the background links do not establish that every Qwen3.5 checkpoint used the same optimizer.
Sources
Every plotted observation links to its evaluation. Reported configurations, benchmark versions and task subsets can differ; similarity describes shared scores, not training lift.
Names used in source tables
Qwen3.5-4B · Qwen3.5-4B (Thinking)