Is MiMo-V2.6-Flash open source?
Yes. The weights are on Hugging Face as XiaomiMiMo/MiMo-V2.6-Flash-RL under the MIT license, published September 21, 2026.
MI
MiMo-V2.6-Flash
Multimodal modelCoding modelMiMo-V2.6-Flash
Also known as: MiMo-V2.6-Flash-RL
Release date: 2026-09-22Views: 97
Parameters
309B
Context length
1M
Multilingual
Supported
Reasoning ability
4/5
MiMo-V2.6-Flash is Xiaomi's efficient omni-modal reasoning model, open-sourced under MIT in September 2026: a 309B sparse MoE with 15B active, 1M context, text/image/video/audio input, and 67.9 on DeepSWE v1.1.
Data sourced primarily from official releases (GitHub, Hugging Face, papers), then benchmark leaderboards, then third-party evaluators. Learn about our data methodology
MiMo-V2.6-Flash
Model basics
Reasoning traces
Supported
Thinking modes
Thinking modes not supported
Context length
1M tokens
Max output length
No data
Model type
Multimodal model
Modality (in / out)
Text, Image, Audio, Video → Text
Release date
2026-09-22
Model file size
No data
MoE architecture
Yes
Total params / Active params
309B / 15B
Knowledge cutoff
No data
MiMo-V2.6-Flash
Open source & experience
Code license
Weights license
MIT License- Commercial use permitted
GitHub repo
N/A
Hugging Face
Live demo
N/A
MiMo-V2.6-Flash
Official resources
Paper
DataLearnerAI blog
N/A
MiMo-V2.6-Flash
API details
API speed
4/5
💡Default unit: $/1M tokens. If vendors use other units, follow their published pricing.
Standard
| Type | Condition | Input | Output |
|---|---|---|---|
| Text | - | ¥1.00/ 1M tokens | ¥2.00/ 1M tokens |
international
| Type | Condition | Input | Output |
|---|---|---|---|
| Text | - | $0.140/ 1M tokens | $0.280/ 1M tokens |
Cache PricingPrompt Cache
| Type | TTL | Write | Read |
|---|---|---|---|
| Text | - | — | $0.0028/ 1M tokens |
| Text | - | — | ¥0.020/ 1M tokens |
“—” means the modality is not billed in that direction, or the vendor has not published a price for it.
MiMo-V2.6-Flash
Benchmark Results
MiMo-V2.6-Flash currently shows benchmark results led by CyberGym (1 / 11, score 95.10), Terminal-Bench 2.1 (18 / 196, score 87.60), AutomationBench (3 / 20, score 52.30). This page also consolidates core specs, context limits, and API pricing so you can evaluate the model from benchmark results and deployment constraints together.
AI Agent - Tool Usage
5 evaluationsBenchmark / mode
Score
Rank/total
Agent Level Benchmark
2 evaluationsBenchmark / mode
Score
Rank/total
MiMo-V2.6-Flash
Publisher
Xiaomi
View publisher details MiMo-V2.6-Flash
Model Overview
MiMo-V2.6-Flash is the efficient omni-modal reasoning model Xiaomi's MiMo team released and open-sourced in September 2026 alongside the flagship MiMo-V2.6-Pro. It targets high-frequency and high-volume workloads at low cost and uses the same mixed reinforcement-learning recipe.
Architecture and specifications
The model is a sparse MoE with 309B total parameters and 15B activated per token, 8 of 256 routed experts active; the language backbone has 48 layers (39 sliding-window attention + 9 global attention) with a hidden size of 4096. The models are natively omni-modal: they accept text, image, video and audio input and return text, with a 1M-token context window. The vision encoder is a 681M-parameter MiMo ViT (28 layers, 24 SWA + 4 full attention); the audio stack combines a 308M AudioTokenizer with a 127M audio patch encoder, and a 5-layer MTP speculative decoder predicts 7 tokens per pass.
Published benchmark results
Xiaomi reports the following for MiMo-V2.6-Flash: DeepSWE v1.1 67.9, Toolathlon-Verified 73.6, Terminal Bench 2.1 87.6, Terminal Bench 4.0 28.8, OSWorld-Verified 80.8, JobBench 61.2, Agents' Last Exam 27.6, AutomationBench v1.0.6 52.3, CyberGym 95.1, MiMo Code Bench 61.2, MiMo VisualCoding 71.5. The official table also compares against Claude Opus 5, GPT-5.6 Sol and Claude Fable 5.
On CyberGym, Flash scores 95.1, slightly above Pro's 94.0.
Pricing
Domestic pricing is ¥1 per million input tokens, ¥0.02 per million cached input tokens and ¥2 per million output tokens; international pricing is $0.14 input, $0.0028 cached input and $0.28 output per million tokens, with a 50% discount for batch inference.
Release and license
The MiMo-V2.6 series was published on Hugging Face on September 21, 2026 under the MIT license, and served through the Xiaomi MiMo platform (mimo.mi.com) from September 22, with distribution also via ModelScope and OpenRouter. Xiaomi published a live dashboard of the RL post-training run: Pro and Flash each took under six days and 30 steps, about 750,000 trajectories in total, at a reported cost of roughly $2.62M and $0.85M respectively.
MiMo-V2.6-Flash
FAQ
How does Flash compare with Pro?
Flash is a 309B sparse MoE with 15B active parameters versus Pro's 1.02T/42B. It trails Pro on most agentic benchmarks (67.9 vs 71.9 on DeepSWE v1.1) but scores slightly higher on CyberGym (95.1 vs 94.0) at roughly a third of the price.
How much does the MiMo-V2.6-Flash API cost?
International pricing is $0.14 per million input tokens, $0.0028 per million cached input tokens and $0.28 per million output tokens.
DataLearner on WeChat
Follow DataLearner on WeChat for AI model updates and research notes.
