MiMo-V2.6-Pro

Xiaomi positions MiMo-V2.6-Pro as its most powerful flagship reasoning model: omni-modal, ultra-high performance, and trillion-parameter, aimed at complex projects, long-horizon tasks, high-stakes work, cybersecurity, and research. The model sits at the top of Xiaomi's MiMo family and was released alongside a full-modality, lower-cost sibling (mimo-v2.6-flash) and a latency-tuned variant (mimo-v2.6-pro-ultraspeed) that promises flagship V2.6-Pro performance at up to 20x faster for real-time workloads.

On third-party benchmarking, MiMo-V2.6-Pro debuted as the top-performing open-weight model on Artificial Analysis' Intelligence Index with a score of 46, surpassing proprietary peers such as xAI's Grok 4.6 (44) and Google's Gemini 3.8 Flash (41). As an open-weight release, the model is notable for pairing that leading benchmark standing with availability for downstream research and product work, though specific training data, safety process, and pipeline details have not been disclosed in available sources.

DevPass (LLM Gateway)mimo-v2.6-promimo

Open provider page Provider docs

Quick Info

Provider: DevPass (LLM Gateway)
Model key: mimo-v2.6-pro
Release date: Sep 22, 2026
Last updated: Sep 22, 2026
Input modalities
Output modalities
Capabilities

Cost

Input token cost: $0.435
Output token cost: $0.87

Limits

Output tokens: 131,072 tokens
Context window: 1,048,576 tokens

Transparent token rates

Compare MiMo-V2.6-Pro pricing

Rates are shown per one million tokens. Combined means one million input plus one million output tokens.

Browse this family

Latest news about MiMo-V2.6-Pro

DevPass (LLM Gateway)

CoverageBenchmark

The eesel article details the MiMo-V2.6-Pro architecture as derived from the Hugging Face model card: 70 layers interleaving local sliding-window attention (window 128) with global attention, 384 routed experts of which 8 fire per token, a 681M-parameter vision encoder, dedicated audio encoders, and a 5-layer multi-tok The piece reports that MiMo-V2.6 was released on September 22, 2026, as an open-weights family from Xiaomi's MiMo team, with Pro as the flagship 1-trillion-parameter sparse MoE. It describes how independent reviewers can obtain weights and an API to run the model, framing MiMo V2.6 as a step on Xiaomi's stated recursiv

DevPass (LLM Gateway)

CoverageBenchmark

Xiaomi's MiMo-V2.6-Pro, released on 22 September 2026, is described as the top open-weights model in the family, scoring 46 on the Artificial Analysis Intelligence Index at roughly US$0.13 per task and sitting one point behind OpenAI's GPT-5.6 Sol on the same index. According to the Flowtivity article, Pro is a sparse The Flowtivity explainer also names the MiMo-V2.6 lineup in three variants — Pro, Flash (309B total / 15B active), and an UltraSpeed latency-optimised build of the Pro checkpoint — and notes the family is released under an MIT license, emphasizing open-weights reuse for businesses. It positions MiMo-V2.6-Pro against Cl