Xiaomi MiMo V2.6 Pro UltraSpeed launches on OpenRouter

What was accelerated in MiMo V2.6 Pro

The specific number that stands out is OpenRouter’s claim of roughly tenfold faster output than the standard MiMo V2.6 Pro. On its model page, OpenRouter presents the release as a speed-optimized edition of the same 1T checkpoint while retaining the quality of the original Pro version.

MiMo V2.6 Pro UltraSpeed appeared on the platform on September 21, 2026. Its listed context window is 1,048,576 tokens, so the speed increase does not appear to come from an obvious reduction in available context. For long agent sessions and work with large repositories, that combination is more compelling than a high tokens-per-second figure alone.

At launch, input cost $4.35 per million tokens and output cost $8.70. Xiaomi’s provider card on the platform also showed throughput of 169 tokens per second and latency of 1.29 seconds. Those are useful concrete reference points, although they should not be projected directly onto every workload.

The first things I would test are not a flattering peak figure, but sustained speed during long generation, time to first token, and behavior with a filled context window. OpenRouter’s metric reflects observed performance for a particular provider listing, not a guaranteed result for every request. Queues, prompt length, and decoding behavior can all change the picture significantly.

There is also a practical OpenRouter advantage: the model is available through the familiar compatible API under the identifier xiaomi/mimo-v2.6-pro-ultraspeed. You do not need to replace an entire client stack just to test a new model.

Where this level of speed actually changes the equation

The biggest gains go to interactive systems where a user or agent needs the next response immediately. Faster output shortens pauses in planning loops, code generation, and sequential tool calls, while the large context window makes it possible to retain working state for longer.

UltraSpeed is not automatically the best choice for batch processing, however. At high output volumes, 169 tokens per second is only one factor; total cost, provider stability, and quality on the actual target tasks matter too. I treat the claim of matching Pro-version quality as a hypothesis for comparative evaluation, not as a universal guarantee.

This is more than a cosmetic Fast-style suffix: the combination of claimed acceleration, long context, and transparent pricing makes the model a genuine option for low-latency inference. The main open question is how well these figures hold up on long requests and under real competition for compute.

We previously covered Pony Alpha’s availability on OpenRouter, including its 200K context window and practical testing use cases. That comparison provides useful context for evaluating what Xiaomi Mimo V2.6 Pro Ultraspeed adds to the platform.