Xiaomi MiMo-V2.6 Distill 9B Report - SWE-bench Pro: leaderboard
Scale AI's SWE-bench Pro: longer, harder issue-resolution tasks than SWE-bench Verified, from public and commercial repositories in several languages (the report does not name the split). From Table 6 of Xiaomi's MiMo-V2.6 technical report, reproduced on the MiMo-V2.6-Distill-Qwen-9B model card: Xiaomi's evaluation of Qwen3.5-9B and of MiMo-V2.6-Distill-Qwen-9B, the released checkpoint made by supervised fine-tuning of Qwen3.5-9B on MiMo-generated data. The report states no agent harness or reasoning setting for this table; its RL column (unreleased per-domain GRPO checkpoints) is not admitted. Score averaged over three runs (avg@3, %), higher is better.
Metric: Score (avg@3, mean of three runs, %). Source: huggingface.co. Saturation forecast: Rough model projection: around 2026. 2 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | MiMo-V2.6-Distill-Qwen-9B (Benchmark Sweep Run: source=MiMo-V2.6 technical report Table 6; evaluated by Xiaomi; harness=unreported; reasoning=unreported; released SFT checkpoint) | 44.6 | |
| 2 | Qwen 3.5 9B (Benchmark Sweep Run: source=MiMo-V2.6 technical report Table 6; evaluated by Xiaomi; harness=unreported; reasoning=unreported; base model) | 32 |
Interactive version: theaggregate.ai/benchmark?slug=xiaomi-mimo-v2-6-distill-9b-report-swe-bench-pro · How It Works · Data refreshed daily, snapshot 2026-11-12.