AITOPAI热报
ModelsMulti-sourceSelected78°

Xiaomi's MiMo-V2.6 RL training

OpenRouter@OpenRouterAI
Post· 4 sources
View the post
Why we picked it

MiMo-V2.6 demonstrates massive RL scaling with 2B tokens per step, offering insights for developers working on large-scale AI training.

Xiaomi live-streamed the reinforcement learning run behind MiMo-V2.6, processing ~2B tokens per step with 1568 prompts and 16 rollouts. The system uses fully async, multi-task agentic RL across mixed harnesses. The company spent nearly half a year studying how far RL can scale. Details will be open-sourced over the coming weeks.

Original · OpenRouter

Xiaomi live-streamed the RL run behind V2.6 while it trained: ~2B tokens per step, 1568 prompts x 16 rollouts, fully async, multi-task agentic RL across mixed harnesses. Details are being open-sourced over the coming weeks. @_LuoFuli on how they scaled it: x.com/_LuoFuli/statu… Fuli Luo @_LuoFuli Nearly half a year of silence. We spent it studying one problem: how far RL can scale. MiMo-V2.6 is in the middle of its RL run right now. Three things we scaled: compute (~2B tokens per step, 1568 prompts × 16 rollouts, fully async), environments and harnesses (multi-task agentic RL, mixed across multiple harnesses in one run), and grader compute (agentic in-group credit assignment, with test-case and rubric-based rewards). We'll open-source the details piece by piece over the coming weeks. Streaming the run: mimo.xiaomi.com/rl/ 🔗 View Quoted Tweet 💬 0 🔄 1 ❤️ 7 👀 1899 📊 1 ⚡

#MiMo-V2.6#Xiaomi#reinforcement learning#RL scaling#agentic RL

Keep reading

Recent stories with the same tag or section
ModelsUS vs ChinaMulti-sourceSelected02:22
Gemini 4 Carbon leads cost scaling in Agent Arena data

Cost differences widen precisely on the longest sessions, so the cheapest model for simple tasks is not the cheapest for complex agentic work. Teams running long-horizon agent tasks such as app development can cut bills by re-benchmarking at high task complexity rather than averaging across workloads. The finding also opens the door for Gemini 4 Argon and Xiaomi's MiMo-V2.6 Flash and Pro to challenge OpenAI and Anthropic where budgets are tightest.

#Gemini 4 Carbon#GPT-6#Claude Opus 5.5#MiMo-V2.6
Event
lmarena.ai@lmarena_ai11 sourcesSource
ModelsUS vs ChinaMulti-sourceSelected73°08:29
Xiaomi releases MiMo-V2.6 multimodal model series

The open publication of real-time training dashboards and technical reports is what sets this release apart, setting a transparency bar few open-source model teams currently meet. Developers tracking open-weight multimodal models should benchmark MiMo-V2.6 against Qwen and DeepSeek releases, as Xiaomi's hardware-software integration could shape public-environment deployments across Asia.

#Xiaomi#MiMo-V2.6#open-source models#multimodal
Event
SuperTechFans4 sourcesSource