Qwopus3.5 9B v3 GGUF EU AI Act Compliance Profile

🌟 Qwopus3.5-9B-v3

💡 Model Introduction

Qwopus3.5-9B-v3 is a reasoning-enhanced model based on Qwen3.5-9B. Its core objective is to simultaneously improve reasoning stability and correctness while optimizing inference efficiency, ultimately achieving stronger cross-task generalization capabilities—particularly in programming.

By continuing to optimize the fundamental structure of its reasoning process alongside high-quality reasoning distillation and structural alignment, it enables the model to achieve higher accuracy rates through shorter, more stable reasoning paths.

🍎 Qwopus3.5-9B-v3: Humaneval Benchmark Evaluation

Inference for models was conducted under the Unsloth runtime environment using bfloat16 (BF16) precision, which provides a balance of numerical range and memory efficiency well-suited to 9B-scale inference. Answer verification, partial chain-of-thought adjudication, and statistical analysis were cross-validated using GPT-4.5-Pro (Thinking) and Claude Opus 4.6 (Thinking) to ensure accuracy and reproducibility of the evaluation outcomes.

HumanEval
I evaluated three 9B-scale Qwen-family models on the full 164-task HumanEval benchmark under a task-level adjudication protocol that resolves code-extraction pollution, answer/code separation issues, and clearly inferable truncated outputs using raw generations. Under this fair and strict evaluation setting, Qwopus3.5-9B-v3 achieves the best base pass@1 of 87.80% (144/164), outperforming both Qwen3.5-9B (82.93%, 136/164) and Claude-Distilled-v2 (82.32%, 135/164). Furthermore, on the stricter plus pass@1 evaluation, Qwopus3.5-9B-v3 also extends its lead to 82.93% (136/164) compared to 77.44% (127/164) for the official baseline (+5.49 pp) and 78.66% (129/164) for the distilled variant.

| Model | Base pass@1 | Plus pass@1 | Rescues (From GPT) | Improvement vs Qwen3.5-9B | |---|---|---|---|---| | Qwopus3.5-9B-v3 | 87.80% (144/164) | *

Your obligations as a Qwopus3.5 9B v3 GGUF deployer

Track your progress

Sign up to track obligations, receive deadline reminders, and generate audit packages.

Tags

Languages

Base Model