Update: MiMo-V2.6-Flash — $0.14/M, 1M-context open omnimodal model
Xiaomi's cheap MiMo tier is a 309B/15B-active MoE — MIT, omnimodal, self-hostable. Pro topped the open leaderboard; Grok reaches Copilot and Bedrock.
Copy markdown
$0.14 in, $0.28 out — or run it on your own GPUs
Flash is a 309B-parameter MoE with only 15B active per token, released Sep 21 under an MIT license. That's flagship-class breadth at $0.14/$0.28 per million tokens on OpenRouter, or free if you self-host the open weights.
A million-token window, and it takes video and audio
The model accepts text, image, video, and audio input across a 1,048,576-token context, and is tuned for coding, visual tasks, and computer use. Build angle: a cheap agent that watches a screen recording or ingests a whole repo in one pass.
It's the budget cut of the top open-source model
Its sibling MiMo-V2.6-Pro scored 46.32 on the Artificial Analysis Intelligence Index v4.3 — the highest open-source result — with Xiaomi claiming parity with Claude Opus 5 and GPT-5.6 Sol on agent benchmarks. Flash is the same family at roughly a third of Pro's $0.435/$0.87 price.
Elsewhere: Grok 4.7 lands in Copilot, 4.6 on Bedrock
xAI's models reached two major builder platforms this week — Grok 4.7 is now selectable in GitHub Copilot for agentic coding, and Grok 4.6 went live on Amazon Bedrock, both at the familiar $2/$6 pricing.