ARVQ-FP4/NVFP4 hybrid (jarrelscy)
source ↗Perplexity (teacher → student)
10.27+1.47baseline 8.8
MiMo-V2.6-Pro
- Organization
- Xiaomi
- Parameters
- 1.02T total / 42B active
- Context
- 1M
- License
- MIT
- Release
- 2026-10
sparse MoE, omni
MiMo-V2.6-Pro-RL is Xiaomi's 1.02T-total / 42B-active omni sparse MoE with a 1M context, released in October 2026 under the MIT license. Its quantization story is a community GGUF and MLX ladder spanning Q2_K 373 GB to Q4_K_M 618 GB, plus one experimental ARVQ-FP4/NVFP4 hybrid that reports a teacher-to-student perplexity table.
MiMo-V2.6-Pro is a 1.02T total / 42B active MIT model from Xiaomi with 1M context. Quantization loss is not measured here — no published per-quant delta shows a quality drop versus the full-precision baseline.
Base-model benchmarks
DeepSWE v1.1
71.9ProgramBench
26.5MiMo Code Bench
63.2AutomationBench
53.1Toolathlon-Verified
76.9Agents' Last Exam
31.6Terminal-Bench 4.0
34.9Terminal-Bench 2.1
89.9OSWorld-Verified
82JobBench
62CyberGym
94
Quant checkpoints
| Checkpoint | Family | Size | Source |
|---|---|---|---|
| XiaomiMiMo MiMo-V2.6-Pro (official) Official Pro checkpoint | BF16 | — | source ↗ |
| ggml-org MiMo-V2.6-Pro-RL MXFP4 | MXFP4 | 554 GB | source ↗ |
| ggml-org MiMo-V2.6-Pro-RL Q2_K | GGUF | 419 GB | source ↗ |
| DevQuasar MiMo-V2.6-Pro-RL Q2_K | GGUF | 373 GB | source ↗ |
| DevQuasar MiMo-V2.6-Pro-RL Q4_K_M | GGUF | 618 GB | source ↗ |
| DevQuasar MiMo-V2.6-Pro-RL Q5_K_M | GGUF | 489 GB | source ↗ |
| mlx-community MiMo-V2.6-Pro-RL mxfp4-q8 mxfp4-q8 | MXFP4 | 516 GB | source ↗ |
Per-quant quality deltas
Serving metrics
No serving metrics captured for this model.
What’s not published
- No official Pro quant checkpoint is published — the ladder is community-built.
- No per-quant quality table beyond the ARVQ hybrid perplexity and KL numbers.
Sources
- https://huggingface.co/XiaomiMiMo/MiMo-V2.6-Flash-RL ↗
- https://huggingface.co/XiaomiMiMo ↗
- https://huggingface.co/XiaomiMiMo/MiMo-V2.6-Pro-RL ↗
- https://huggingface.co/ggml-org/MiMo-V2.6-Pro-RL-GGUF ↗
- https://huggingface.co/DevQuasar/XiaomiMiMo.MiMo-V2.6-Pro-RL-GGUF ↗
- https://huggingface.co/mlx-community ↗
- https://huggingface.co/jarrelscy/MiMo-V2.6-Pro-RL-ARVQ-hybrid ↗
FAQ
- What is MiMo-V2.6-Pro-RL's license, context window and parameter count?
- MIT license, 1M context window, 1.02T total / 42B active parameters (Xiaomi).
- How much quality does quantization cost MiMo-V2.6-Pro-RL?
- Across 1 published delta, quantized checkpoints differ from the full-precision baseline by 1.47 points on average (range 1.47 to 1.47).
- How many quant checkpoints are cataloged for MiMo-V2.6-Pro-RL?
- 7 checkpoints across BF16, MXFP4, GGUF, sizes from 373 GB to 618 GB.
- When was MiMo-V2.6-Pro-RL released?
- 2026-10.
- Where do the MiMo-V2.6-Pro-RL benchmark numbers come from?
- From 7 official sources — vendor model cards, quantization publisher cards and serving recipe pages.