MiMo v2.6 Flash
MiMo-V2.6-Flash is an open-source foundation model from Xiaomi, built on a Mixture-of-Experts architecture with 309B total parameters and 15B activated per token. It uses a hybrid attention mechanism, accepts text, image, audio and video, and costs $0.14 per million input tokens.
Technical Specifications
Live data fetched from the OpenRouter model registry — always up to date.
What is Xiaomi: MiMo-V2.6-Flash?
MiMo-V2.6-Flash is Xiaomi's open-source foundation model. Its Mixture-of-Experts design activates 15B of 309B parameters per token, and a hybrid attention mechanism keeps long-context inference efficient. It is omni-modal - text, images, audio and video - and it comes in at $0.14 per million input tokens with a 1,050,000-token context window.
Features & Capabilities
309B MoE
309B total parameters with only 15B active per token.
1.05M Context
Just over a million tokens of working memory.
Text, Image, Audio, Video
Omni-modal input in an open-weight model.
Hybrid Attention
Keeps long-context inference efficient rather than quadratic.
Open Source
Weights and architecture are publicly documented.
$0.14 per Million
Near the cheapest omni-modal option on the market.
Who Is Xiaomi: MiMo-V2.6-Flash For?
Self-Hosted Evaluation
Open weights make on-premise testing realistic.
Media Pipelines
Audio and video understanding at very low cost.
Long Document Work
Over a million tokens of source material.
High-Volume Agents
Many calls at a price that permits them.
Xiaomi: MiMo-V2.6-Flash Pricing
- $0.14 per million input tokens, $0.28 per million output tokens
- Open-source model - weights are publicly available
- Run it from your Hoodgen free credits, no credit card required
- Higher allowances on Hoodgen AI paid plans
What Makes Xiaomi: MiMo-V2.6-Flash Different?
Xiaomi entering the model market properly is the interesting part here. MiMo-V2.6-Flash is not a branding exercise: 309B parameters with 15B active, a hybrid attention mechanism for long context, full omni-modal input, and open weights. For a first serious frontier-adjacent release that is a strong package.
The hybrid attention design is what makes the 1.05M-token window practical. Standard attention becomes expensive quickly as context grows; mixing sparse and linear attention is the standard answer and it is applied here. The result is long-context ability without the inference cost that usually accompanies it.
For teams that need to run on their own hardware, the open weights matter more than the hosted price. Hoodgen exposes the hosted version so you can evaluate quality and cost first, then decide whether to self-host. Either way, $0.14 per million input tokens for omni-modal work is close to the floor.
Xiaomi: MiMo-V2.6-Flash Performance & Pricing
MiMo-V2.6-Flash is the cheapest omni-modal model in the current Xiaomi line and one of the cheapest anywhere.
What Can You Do With Xiaomi: MiMo-V2.6-Flash?
MiMo-V2.6-Flash is for work that has to run cheaply over media or very long documents. It is also the sensible first stop if you are considering self-hosting an omni-modal model.
Best Use Cases
- Bulk audio and video processing
- Long-document analysis over a million tokens
- Evaluating an open-weight omni-modal model
- High-volume agent pipelines
- Cost benchmarking against hosted alternatives
Xiaomi: MiMo-V2.6-Flash Privacy & Data Policy
MiMo-V2.6-Flash is developed by Xiaomi and served through OpenRouter. The model weights are open source, so self-hosting is also an option.
If data residency matters, the open weights mean you can run the model on your own hardware rather than sending material to a hosted provider. Hoodgen exposes the hosted version for evaluation.
Xiaomi: MiMo-V2.6-Flash vs Other Models
Frequently Asked Questions
Open-source omni-modal on a budget
Chat with MiMo-V2.6-Flash on Hoodgen - sign in and start in seconds.
Chat with Xiaomi: MiMo-V2.6-Flash