Xiaomi's MiMo-V2.6-Pro scores well above median on Intelligence Index
Artificial Analysis published an intelligence, performance and price analysis of MiMo-V2.6-Pro, an open weight reasoning model that Xiaomi released on September 21, 2026. The model scored 46 on the Artificial Analysis Intelligence Index, a composite benchmark covering reasoning, knowledge, mathematics and coding, placing it well above the median of 18 among comparable open weight models of similar size.
MiMo-V2.6-Pro is a Mixture-of-Experts model with 1.0 trillion total parameters and 42 billion active parameters, released under the MIT license so its weights can be downloaded and self-hosted. It handles text, image, speech and video input and produces text output, with a context window of 1.0 million tokens. It is a reasoning model: it uses extended thinking, or chain-of-thought, before answering.
On speed, the analysis measured 124.5 tokens per second through Xiaomi's own API (a rounded 125 tokens per second in the summary), well above the 74.8 tokens per second median for open weight models of similar size. Time to first token came in at 2.34 seconds, which the analysis labels better than average against a median of 2.33 seconds, though the two figures are close.
Pricing is $0.43 per 1 million input tokens and $0.87 per 1 million output tokens through Xiaomi's API; blended at a 7:2:1 cache-hit-to-input-to-output ratio, that works out to $0.18 per 1 million tokens. Running the full Intelligence Index evaluation cost $206.66 and produced 140 million output tokens, matching the median for comparable models even though the analysis still calls the model somewhat verbose. The analysis is not internally consistent on how that pricing compares to the field: its opening summary puts the median input price at $0.30 and the median output price at $1.13, calling MiMo-V2.6-Pro's input price somewhat expensive; its FAQ section instead gives medians of $0.45 and $1.68 for what appears to be a differently scoped comparison set, and calls the same input price better than average. The page does not reconcile the two figures.
MiMo-V2.6-Pro is currently available through one API provider. The analysis names no specific competitor models: all of its comparisons are against an unnamed median of open weight or price-comparable models.
Key facts
- MiMo-V2.6-Pro scores 46 on Artificial Analysis's Intelligence Index, well above the median of 18 among comparable open weight models.
- It runs at about 124.5 to 125 tokens per second, well above the 74.8 t/s median for open weight models of similar size.
- Pricing is $0.43 per 1M input tokens and $0.87 per 1M output tokens, but the analysis gives two different median comparisons for that same pricing ($0.30/$1.13 in its summary, $0.45/$1.68 in its FAQ) without reconciling them.
- It is a Mixture-of-Experts reasoning model with 1.0 trillion total parameters (42 billion active), a 1M token context window, and an MIT license; Xiaomi released it on September 21, 2026.
- It accepts text, image, speech and video input and returns text, but is currently available through only one API provider.
Why it matters
MiMo-V2.6-Pro adds another strong open weight entrant to the reasoning-model field: a score of 46 on the Intelligence Index against a field median of 18 is a wide margin, and the model gets there while also running fast and pricing close to the middle of the market. Combined with a 1 million token context window, multimodal input (text, image, speech and video) and an MIT license that permits commercial self-hosting, it gives Xiaomi a genuinely competitive open weight release rather than a token one.
Who it affects
Teams choosing an open weight model for self-hosting or via API, particularly ones that need reasoning ability, a long context window or multimodal input, gain another option to weigh against other open weight and proprietary models in the same size and price class. Xiaomi's rivals among open weight model makers now have a new benchmark to match.
How to use it
MiMo-V2.6-Pro is open weights under the MIT license, so it can be downloaded and self-hosted, and it is also available through one API provider, Xiaomi's own, priced at $0.43 per 1 million input tokens and $0.87 per 1 million output tokens, or $0.18 per 1 million tokens blended at a 7:2:1 cache-hit-to-input-to-output ratio. It accepts text, image, speech and video input and returns text, with a 1 million token context window.
How solid is it
The figures come from Artificial Analysis, a third-party benchmark and pricing tracker, not from Xiaomi's own marketing, and the write-up covers intelligence score, speed, latency, cost and token verbosity rather than a single headline number. That said, the page itself is not fully consistent: it states two different median input and output prices for the same comparison, $0.30 and $1.13 in its opening summary versus $0.45 and $1.68 in its FAQ, without explaining why they differ or reconciling them, and it labels MiMo-V2.6-Pro's identical $0.43/$0.87 pricing as somewhat expensive in one place and better than average in the other. It also calls a 2.34 second time to first token better than average against a median of 2.33 seconds, a gap too small to read much into.
Risks and caveats
The model is available through only one API provider right now, which limits how well price and reliability can be checked against alternatives. The analysis does not name which specific models make up its median comparisons, so it is hard to independently verify where MiMo-V2.6-Pro actually lands against named competitors. It also generated 140 million output tokens on the Intelligence Index run, the same as the median but still flagged as somewhat verbose, meaning real costs could run higher than the headline per-token price suggests on verbose tasks.