# Xiaomi MiMo V2.6 Pro and Flash: open-weights omnimodal flagships trained live in public

> Xiaomi open-sourced MiMo V2.6 Pro (1T) and Flash (309B) on Sept 21, 2026: 1M omnimodal context, RL trained live in public, AA Index 46.32, DeepSWE 72.6%. Pro vs Flash vs UltraSpeed prices, benchmarks and local-AI takeaways.

*Source: https://velstech.net/mimo-v2-6 · Updated: 2026-09-22 · Category: AI · Tags: LLM, Xiaomi, MiMo, Benchmarks, Open Weights, AI News*

*Markdown version of [Xiaomi MiMo V2.6 Pro and Flash: open-weights omnimodal flagships trained live in public](https://velstech.net/mimo-v2-6). [Read the full guide with interactive tools](https://velstech.net/mimo-v2-6).*
*Also as Markdown: [Hindi](https://velstech.net/mimo-v2-6.hi.md) · [Tamil](https://velstech.net/mimo-v2-6.ta.md).*

---

Xiaomi did two unusual things on September 21, 2026. It released **MiMo-V2.6
Pro** and **MiMo-V2.6 Flash** – two natively omnimodal models,
open-sourced under the MIT license – and it trained them *live in public*,
with a dashboard showing costs, token throughput, and benchmark scores as the RL
run progressed. No US frontier lab does this. I pieced together the launch
announcement, the API docs, the training livestream reporting, and the third-party
benchmark trackers so you get one honest picture: what Pro, Flash, and the
UltraSpeed variant each are, what the numbers actually say, and what open weights
at this scale mean for people running local AI.

This guide covers the full V2.6 family – Pro vs Flash vs Pro-UltraSpeed on specs,
price, and benchmarks – plus the training story, the "Vibe World" demos, and the
practical question of what you can actually run yourself.

*Source note:* specs and prices below come from Xiaomi's
MiMo-V2.6 announcement, the [MiMo API
model docs](https://mimo.mi.com/docs/en-US/quick-start/summary/model), and gateway listings (Vercel AI Gateway, OpenRouter). Scores are a
mix of Xiaomi-reported RL curves and third-party trackers (Artificial Analysis,
LLM Stats, Benchable) – labelled each time, because the two kinds of numbers mean
different things.

## The short version

- Three flavours, one family: MiMo-V2.6-Pro (flagship, ~1T
parameters), MiMo-V2.6-Flash (309B total / 15B active MoE, the efficiency pick),
and Pro-UltraSpeed (same quality at up to 20x output speed – 1,000 tok/s peak –
for latency-sensitive work at 10x the price).

- Natively omnimodal: text, image, video, and audio in – with a
1M-token context window (1,048,576) and up to 128K output – plus deep thinking,
tool calling, JSON mode, and prompt caching.

- Top open-weights scores: Pro hits 46.32 on the Artificial
Analysis Intelligence Index – Xiaomi's claimed highest open-source result, ahead
of Kimi K3 and Qwen3.8 Max – and DeepSWE v1.1 climbs from 58.4 to 72.57 during
the RL run (Flash: 48.8 → 65.68).

- Pricing holds the V2.5 line: Flash $0.14 input / $0.28 output per
million, Pro $0.435 / $0.87, with near-free cache reads ($0.0028 / $0.0036).
UltraSpeed costs 10x Pro.

- MIT-licensed open weights on Hugging Face, plus the technical
report, training environments, and RL code – with live availability in AI Studio,
MiMo Code, MiMo Desktop, the MiMo API, OpenRouter, and Vercel AI Gateway.

## What each variant is

| Spec | MiMo-V2.6-Pro | MiMo-V2.6-Flash | Pro-UltraSpeed |
| --- | --- | --- | --- |
| Total parameters | ~1T (reported ~1.02T) | 309B | Same as Pro |
| Active per token | Not disclosed (MoE) | 15B (hybrid attention MoE) | Same as Pro |
| Context / max output | 1,048,576 / 128–131K | 1,048,576 / 128–131K | Same as Pro |
| Input modalities | Text, image, audio, video, PDF | Text, image, audio, video, PDF | Same as Pro |
| Output | Text | Text | Text, up to ~20x faster |
| Speed tech | Standard serving | Efficient MoE | FP4 lossless quant + DFlash parallel decoding + TileRT kernels |
| Input / output per 1M | $0.435 / $0.87 | $0.14 / $0.28 | 10x Pro |
| Cached input per 1M | $0.0036 | $0.0028 | – |
| License | MIT (open weights) | MIT (open weights) | Hosted only |

The positioning is deliberate: Pro pushes the capability ceiling (Xiaomi says on
par with Claude Opus 5 and GPT-5.6 Sol across most agent benchmarks), Flash holds
most of the intelligence at roughly a third of the price, and UltraSpeed trades
money for latency – up to 1,000 tokens/s – for realtime coding agents. Note the
cache-read prices: at ~1/50th (Flash) to ~1/120th (Pro) of the miss rate, agents
that reuse context – which is nearly all of them – cost far less than sticker.
That is the same trick that makes [DeepSeek V4.1
Flash](https://velstech.net/deepseek-v4-1-flash) so cheap in practice.

## Trained in public: the RL livestream

This is the part of the release I find most important, more than any single score.
Led by Luo Fuli – the ex-DeepSeek scientist Lei Jun hired to run the MiMo team,
who previously led DeepSeek-V2 work – Xiaomi spent roughly half a year studying one
problem ("how far RL can scale"), then ran the V2.6 RL phase in the open: a live
dashboard with training progress, token consumption, cost meters, step timings,
dataset composition across 23 categories, and benchmark curves updating mid-run.

The run itself: roughly 30 RL steps each for Pro and Flash across ~750,000
trajectories in under six days, on coding, general-agent, visual, and
cybersecurity tasks, with 1,568 samples per update and contexts up to 1M tokens.
Xiaomi says it froze the router to limit drift and used adversarial evaluation,
anomaly detection, and verifier cross-checks against reward hacking. Final RL
costs reported at ~$2.62M for Pro and ~$0.85M for Flash (a mid-run September 17
snapshot showed ~$512K at 49.4B Flash tokens – the meter kept running after that).
Mid-training, Pro had already jumped from the V2.5 baseline of 19% to 65.97% on
DeepSWE v1.1, sitting in a tier with Grok 4.6 (67%), Kimi K3 (69%), and Fable 5
(70%) – and the final numbers went higher still (see below).

Why this matters beyond Xiaomi: closed-door RL is the industry norm, which means
nobody can check whether a capability jump came from the algorithm, the data, or
the eval harness. A public cost-and-curve ledger – $128K-a-day-scale spending
visible to everyone – is a reproducibility experiment as much as a marketing stunt.
If you train your own models, the published environments and RL code are arguably
the most valuable artifact of this release.

## Benchmarks: where Pro leads and where it doesn't

Xiaomi's headline: **46.32 on the Artificial Analysis Intelligence Index
v4.3** – its claimed highest open-source score, ahead of Kimi K3 and
Qwen3.8 Max. Vendor framing aside, the RL curves are striking: DeepSWE v1.1 Pro
58.4 → **72.57**, Flash 48.8 → **65.68**. Third-party
trackers add texture – Flash's reported strengths cluster in agentic and security
work:

| Benchmark | MiMo-V2.6-Flash | MiMo-V2.6-Pro |
| --- | --- | --- |
| CyberGym | 95.1% | – |
| Terminal-Bench 2.1 | 87.6% | – |
| OSWorld-Verified (computer use) | 80.8% | – |
| MiMo Cyber Bench | 77.2% | – |
| Toolathlon-Verified | 73.6% | – |
| DeepSWE v1.1 | 65.68% (RL final) | 72.57% (RL final) |
| MiMo Coding Bench | – | 63.2% |
| ProgramBench | – | 26.5% |

The honest read: this is the second open-weights family in a month – after
[DeepSeek V4.1 Flash](https://velstech.net/deepseek-v4-1-flash) – that looks genuinely
competitive with closed flagships on *agent* work (terminal, computer use,
repo coding) rather than just chat. The gaps are where you'd expect for
efficiency-tuned open models: closed-book reasoning and the newest harnesses still
belong to [Astra](https://velstech.net/gpt-6-astra) and
[Fable 5.1](https://velstech.net/claude-fable-5-1). And cross-vendor score comparisons
are directional – different harnesses, different effort settings, different dates.
LLM Stats' head-to-head composite currently has Flash and Pro closely matched
(45.7 vs 40.6), with Pro winning the three directly shared benchmarks.

## Beyond chat: "Vibe World," robots, and real science demos

Xiaomi is pushing V2.6 past conventional software work into what it calls
**"Vibe World"**: from an image, video, or text prompt, the model
coordinates agents to build and visually test interactive 3D scenes – Blender
assets, frontends and presentations, assembled video, even music as scores and
MIDI. The wilder demo: controlling a Franka Panda robotic arm from camera feeds.
Research demos included screening materials for capturing PFAS "forever chemicals"
and helping formalize a Lean 4 theorem in 6,000+ lines of kernel-verified code.

Treat demos as demos – carefully chosen, possibly retried – but the direction is
consistent with the benchmarks: this family is built for agents that *do
things* (browse, comprehend, reason, act) across modalities, not for winning
trivia. The 1M native omnimodal context is what makes hour-long multimodal agent
runs architecturally possible.

## Price, access, and how to try it

- API: Flash $0.14 / $0.28 per million input / output, Pro $0.435 / $0.87, UltraSpeed 10x Pro. Model IDs xiaomi/mimo-v2.6-flash and xiaomi/mimo-v2.6-pro on gateways; native endpoints on the MiMo API platform with deep-thinking, tool-calling, JSON mode, and prompt caching.

- Apps: AI Studio, MiMo Code, MiMo Desktop (leaving early access with both models), MiMo Studio chat with no setup.

- Open weights: MIT license on Hugging Face (XiaomiMiMo org) – plus technical report, training environments, and RL code on GitHub. Self-hostable if you have datacenter-class hardware (see below).

- Context: run the price math for your workload in our AI API cost calculator – at these cache-hit discounts, the break-even vs self-hosting moves a lot.

## What this means if you run local AI

First the honest answer: **you are not running 1T – or even 309B – at
home.** Flash at 309B total needs datacenter-class VRAM in full precision;
Pro at ~1T is firmly in the 2,000-GPU-plus-storage deployment world. My 12 GB
RX 6800M will not be downloading either one – see
[how much VRAM LLMs need](https://velstech.net/how-much-vram-for-llm) and the
[local-LLM GPU guide](https://velstech.net/best-gpu-for-local-llm) for what actually fits.

What you *can* use: the MIT weights mean community FP8 originals, [GGUF quants](https://velstech.net/llama-cpp-guide), and vLLM/SGLang support will mature fast – and the API is among the
cheapest frontier-class inference available (Flash at $0.14/$0.28 undercuts even
DeepSeek V4.1 Flash's peak rates on input). The practical setup for local-AI
folks: keep private and offline work on your local 7B–35B models, route heavy
multimodal agent jobs to the $0.14 API, and watch the published RL code – the
training transparency here teaches more than the weights alone. Our
[lab benchmarks](https://velstech.net/benchmarks/index) show exactly where the local
side wins: fast, private, repeatable generation with zero per-token cost.

## Bottom line

MiMo-V2.6 is the most significant open-weights release of the month: a
frontier-competitive agent family (46.32 AA Index, DeepSWE 72.6%) at
open-model prices, MIT-licensed, with its training receipts published live. Flash
is the value default for agent builders; Pro is the ceiling-pusher; UltraSpeed is
for when latency is the product. And the livestreamed RL run sets a precedent I
hope other labs copy – because "trust our curve" is no substitute for watching
the meter run.

## Sources

- Xiaomi MiMo: Introducing MiMo-V2.6 series (Sept 21–22, 2026)

- MiMo API docs: model capabilities, 1M context limits, rate limits, pricing

- TestingCatalog: Xiaomi open-sources MiMo-V2.6 Pro and Flash models

- Forkast News: Xiaomi MiMo-V2.6 trains in public – live RL dashboard

- Artificial Analysis: MiMo-V2.6-Pro model page – Intelligence Index 46.32

- LLM Stats / Benchable / OpenRouter / Vercel AI Gateway: V2.6 specs, benchmarks, pricing

## FAQ

**What is the difference between MiMo V2.6 Pro, Flash, and UltraSpeed?**

Pro is the ~1T-parameter flagship for maximum capability; Flash is a 309B-total / 15B-active MoE holding most of the intelligence at roughly a third of the price ($0.14/$0.28 vs $0.435/$0.87 per million tokens); UltraSpeed serves Pro-quality output up to 20x faster (1,000 tok/s peak) at 10x the price for latency-sensitive work.

**Are MiMo V2.6 models open source?**

Yes. Pro and Flash are released under the MIT license with weights on Hugging Face, plus Xiaomi's technical report, training environments, and RL code on GitHub.

**How good is MiMo V2.6 Pro?**

It scores 46.32 on the Artificial Analysis Intelligence Index – Xiaomi's claimed highest open-source result, ahead of Kimi K3 and Qwen3.8 Max – and DeepSWE v1.1 rose from 58.4 to 72.57 during its publicly livestreamed RL run.

**Can I run MiMo V2.6 locally?**

Not on consumer hardware – 309B (Flash) to ~1T (Pro) parameters need datacenter-class VRAM. Home users get the cheap API ($0.14/MTok input for Flash), the free Studio/Desktop apps, or wait for community quantizations.

---

*VelsTech – technology explained for everyone. Original: https://velstech.net/mimo-v2-6*