Xiaomi open sources MiMo-V2.5 and MiMo-V2.5-Pro under the MIT License, saying both models are among the most efficient available for agentic “claw” tasks
Xiaomi, the Chinese firm best known for its smartphones and electric vehicles, has lately been shipping some incredibly affordable …
VentureBeatCarl Franzen
Context & Ripple Effects
Xiaomi’s MiMo effort has progressed from an open-source reasoning model to MiMo-V2 and now a pair of MIT-licensed V2.5 releases aimed at agentic tasks. Related coverage also points to continued iteration across coding, mixture-of-experts, and high-throughput variants.
The significance is less a single benchmark claim than the licensing choice: Xiaomi is making a newer agent-oriented model family available for downstream use and modification, rather than limiting access to an API.
First-order effects
Developers and enterprises can inspect, adapt, and deploy MiMo-V2.5 and MiMo-V2.5-Pro under MIT terms, subjecting Xiaomi’s efficiency claims to practical evaluation rather than vendor-only access.
Xiaomi expands MiMo from a model announcement into a reusable open-model distribution, strengthening its position among organizations building agentic and coding-oriented AI tooling.
Second-order effects
Model buyers gain another option to test against proprietary APIs and other open-weight systems, increasing pressure on providers to compete on deployment cost, speed, customization, and reliability—not just headline capability.
Tool builders can target a locally deployable MiMo variant for agentic workflows, while infrastructure providers may see demand shift toward serving and optimizing these models.
Third-order effects
If successive MiMo releases remain openly licensed and competitive, more AI procurement will become a portfolio decision between self-hosted adaptable models and managed proprietary services.
The pattern points toward model competition being shaped increasingly by distribution rights, inference efficiency, and ecosystem adoption alongside raw benchmark performance.
The trend: Open licensing of increasingly capable agent-oriented models is widening buyer choice and shifting AI competition toward deployability, efficiency, and downstream ecosystem control.
Just dropped two open-source models: MiMo-V2.5-Pro (Code Agent, 1T total) and MiMo-V2.5 (Multimodal Agent, 310B total). Oh and one more thing — we're giving devs & creators 100T tokens on us. Go build something cool 🛠️ 🎁 100T Free Token Grant for Builders
Xiaomi's MiMo V2.5 Pro has landed at 54 in the Artificial Analysis Intelligence Index, tied with Moonshot's Kimi K2.6 - the current top open weights model. MiMo V2.5 Pro's weights are expected to be released soon, which would make MiMo V2.5 Pro the first equal open weights model …
xiaomi mimo v2.5 eval card, pro is 1T total 42B active, omni (video/image/audio) is 310B total 15B active, both have 1M context support they train in FP8, 27T tokens for pro and 48T for the smaller variant. interleaved SWA with an aggressive 6:1 ratio and 128 window size, still […
SGLang and vLLM support for the MiMo-V2.5 series is here. 🙌 Huge thanks to SGLang project from @lmsysorg and @vllm_project for moving fast and helping developers get started with MiMo-V2.5 on day zero. [image]
New Opensource SoTA contender enters the arena Xiaomi MiMo-V2.5 Pro - 1.02T Total Params / 42B Active Params - Base and Instruct versions Xiaomi MiMo-V2.5 - 310B Total Params / 15B Active Params - Base and Instruct versions MIT License Opensource AI just keeps getting better [ima…
🎉 MiMo-V2.5 series is here, day-0 support is now live in SGLang! Two models to try: 1️⃣ MiMo-V2.5-Pro: 1.02T/42B MoE, hybrid attention, up to 1M context 2️⃣ MiMo-V2.5: full multimodal (text, image, video, audio), 310B/15B MoE, 1M context We also have day 0 support for this model …
MiMo-V2.5 achieved Day-0 adaptation across multiple chip platforms on the first day of open source release. Huge thanks to our hardware ecosystem partners for helping make MiMo-V2.5 easier to deploy and run efficiently across more environments: @awscloud、@AMD [image]
...It wasn't an intern's joke MiMo 2.5 (not Pro): > Trained on a total of ~48T tokens using FP8 mixed precision. The context window supports up to 1M tokens. We've got another 1M class, and the largest disclosed pretrain. Congrats Xiaomi. [image]