RLTasksby ulam.ai
Models / Qwen

Qwen / MODEL EVIDENCE

Qwen3.5-397B-A17B

35 benchmarks · 40 observations · 10 sources · released 2026-02-13

Scores at a glance

35 benchmarks · large dot: Qwen3.5-397B-A17B; small dots: closest models where reported

IFEval ↗92.6
90.994.8

Qwen3.5-397B-A17B · IFEval · Qwen report ↗

BFCL V4 ↗72.9
63.177.5

Qwen3.5-397B-A17B · BFCL-V4 · Qwen report ↗

MCPMark ↗46.1
29.557.5

Qwen3.5-397B-A17B · MCP-Mark · original release · Qwen report ↗

MMMU ↗85.0
80.687.2

Qwen3.5-397B-A17B · MMMU · Qwen report ↗

CharXiv ↗80.8
66.182.1

Qwen3.5-397B-A17B · CharXiv(RQ) · Qwen report ↗

LVBench ↗75.5
57.376.2

Qwen3.5-397B-A17B · LVBench · Qwen report ↗

GPQA Diamond ↗86.36%
86.3686.36

Qwen3.5 397B-A17B · GPQA diamond · Best score (across scorers) · % ↗

1 more reported result

85.86% · Qwen3.5 397B-A17B · GPQA diamond · Best score (across scorers) · % · Epoch AI · GPQA diamond ↗

LMCA ↗37.89%
2.7868.26

Qwen3.5 397B-A17B · LMCA · Score · % ↗

DTBench ↗87.47%
41.6598.93

Qwen3.5 397B-A17B · DTBench · Accuracy · % ↗

ALE-Bench ↗621.92 Performance points
137.782951.32

qwen3.5-397b-a17b · Average contest performance ↗

Closest models by score

Similar positions on shared benchmark leaderboards. Smaller gaps are closer.

32.2 percentile-point gap

25 of 31 shared benchmark comparisons

39.3 percentile-point gap

25 of 31 shared benchmark comparisons

46.8 percentile-point gap

28 of 31 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
Stateful agent training →RL training systems · background →

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.

  1. Qwen3.5-397B-A17B model card ↗

    Qwen Team · 2026

Filter comparisons to this model →

Reports featuring this model

Qwen3.5 · earlier-2026 baseline · Qwen →

Sources

Every plotted observation links to its evaluation. Reported configurations, benchmark versions and task subsets can differ; similarity describes shared scores, not training lift.

  1. DTBench — Conceptual Reasoning Index ↗

    Conceptual Reasoning Index

  2. LMCA — Conceptual Reasoning Index ↗

    Conceptual Reasoning Index

  3. Ale Bench — methodology and published results ↗

    Epoch AI · 2026

  4. APEX-Agents ↗

    Epoch AI

  5. FrontierMath Tiers 1-3 (v2) ↗

    Epoch AI

  6. GPQA Diamond ↗

    Epoch AI

  7. Mystery Game Puzzles ↗

    Epoch AI

  8. OTIS Mock AIME 2024-2025 ↗

    Epoch AI

  9. AI benchmarks & capabilities ↗

    Epoch AI

  10. Qwen/Qwen3.5-397B-A17B · Hugging Face ↗

    Qwen Team

Names used in source tables

Qwen3.5 397B-A17B · Qwen3.5-397B-A17B · qwen3.5-397b-a17b