MiMo-V2.6 Pro UltraSpeed

// Overview

MiMo-V2.6-Pro-UltraSpeed is the third model listed in Xiaomi's MiMo-V2.6 series, alongside MiMo-V2.6 Pro and MiMo-V2.6 Flash. Xiaomi's own model-updates page dates the series 2026-09-22 and describes this entry as "Flagship V2.6-Pro performance, up to 20x faster. Built for real-time and latency-sensitive workloads."

Where Pro and Flash are positioned by capability and cost, UltraSpeed is positioned purely by latency. Its model page states that it "retains V2.6-Pro's full flagship-level capability with zero compromise" while delivering "up to 20x faster output speed than V2.6-Pro". Xiaomi does not document the mechanism behind the speed-up on that page, and publishes no separate benchmark table for it.

The specification otherwise mirrors the flagship: a 1 million-token context window, up to 128K tokens of output, and omnimodal input — text, image, video and audio in, text out. Unlike MiMo-V2.6-Pro-RL and MiMo-V2.6-Flash-RL, which Xiaomi publishes as MIT-licensed checkpoints on Hugging Face, no UltraSpeed checkpoint appears under the XiaomiMiMo organisation: it is reached through the MiMo API platform rather than downloaded.

Pricing is listed per million tokens: $4.35 for input on a cache miss, $0.036 for input on a cache hit, and $8.70 for output (¥30, ¥0.25 and ¥60 respectively).

// Pricing

Cache-miss input $4.35, cache-hit input $0.036, output $8.70 (¥30 / ¥0.25 / ¥60).

Pricing source ↗

// Strengths

• Up to 20x the output speed of MiMo-V2.6 Pro, at what Xiaomi describes as the same flagship-level capability
• The full 1M-token context and 128K-token output ceiling of the flagship, not a shortened variant
• Omnimodal input — text, image, video and audio — in a latency-tuned model
• A cache-hit input price of $0.036 per 1M tokens, roughly 1/120th of the cache-miss rate, for prompt-heavy repeat traffic

// Best for

• Reach for it when an interactive product — voice, live chat, a co-pilot — is bounded by how fast tokens arrive rather than by peak reasoning quality.
• Reach for it for latency-sensitive production paths where MiMo-V2.6 Pro is the right model but its output rate is the bottleneck.
• Reach for it when a long-context pipeline replays a large shared prefix, so most input tokens bill at the cache-hit rate.
• Reach for it for multimodal streams — screen, video or audio in, text out — that have to keep up with the user in real time.

// How to access

// MiMo — every version

The full lineage of the MiMo line, newest first. Every version has its own page — click any to compare specs, benchmarks and pricing.