Xiaomi Lists Three New MiMo-V2.6 Models on OpenRouter: Pro, Flash, and Pro-UltraSpeed
Xiaomi has added three new entries to its MiMo model family on OpenRouter: MiMo-V2.6-Pro, MiMo-V2.6-Flash, and MiMo-V2.6-Pro-UltraSpeed. Full specifications, pricing, and benchmark results have not yet been disclosed.
Xiaomi Lists Three New MiMo-V2.6 Models on OpenRouter
Xiaomi has published three new model listings under its MiMo brand on OpenRouter: MiMo-V2.6-Pro, MiMo-V2.6-Flash, and MiMo-V2.6-Pro-UltraSpeed. The listings confirm the models exist and are accessible through OpenRouter's API, but Xiaomi has not yet released a technical report, benchmark scores, context window specifications, or pricing for any of the three variants.
What we know
All three models appear under the xiaomi/ namespace on OpenRouter, following a naming convention that mirrors patterns used by other frontier labs — a flagship "Pro" tier, a lighter and presumably cheaper "Flash" tier for high-throughput use cases, and an "UltraSpeed" variant of the Pro model apparently optimized for lower latency. This structure suggests Xiaomi is targeting different points on the cost-versus-capability curve with a single generation of models, similar to how Google segments its Gemini line into Pro and Flash tiers.
Beyond the naming, no confirmed technical details are currently available:
- Context window: not yet disclosed
- Pricing (input/output per 1M tokens): not yet disclosed
- Parameter count: not yet disclosed
- Benchmark scores: not yet disclosed
- Training data cutoff: not yet disclosed
Xiaomi's MiMo series first drew attention earlier in 2025 with smaller reasoning-oriented models aimed at coding and math tasks, positioning the company as one of several Chinese hardware manufacturers — alongside Xiaomi's smartphone and IoT rivals — investing in proprietary foundation model research rather than relying solely on third-party models for their software ecosystems. The V2.6 designation indicates this is an iterative update to that line rather than a ground-up architectural change, though Xiaomi has not published a changelog explaining what improved between prior MiMo releases and this version.
Why three variants at once
Releasing Pro, Flash, and a UltraSpeed variant simultaneously is consistent with a strategy of covering multiple deployment scenarios in one launch window: the Pro model for maximum capability, Flash for cost-sensitive or high-volume applications, and Pro-UltraSpeed for latency-critical use cases such as real-time assistants or on-device-adjacent applications tied to Xiaomi's hardware ecosystem. Without published benchmarks, it is not possible to independently verify whether the UltraSpeed variant trades accuracy for speed, or whether Flash represents a distilled or otherwise smaller version of Pro.
What this means
This release is notable primarily as a signal of intent: Xiaomi is expanding its MiMo lineup into a multi-tier structure that mirrors the segmentation strategies of Google, Anthropic, and OpenAI. But with no benchmark data, pricing, or architectural details public yet, it is too early to assess where these models sit competitively against established options like GPT, Gemini, Claude, or Chinese rivals such as Qwen and DeepSeek. Developers evaluating these models should treat the OpenRouter listings as an early access point rather than a full product launch, and watch for a technical report or benchmark disclosure from Xiaomi to confirm actual performance claims.
Related Articles
Z.ai founder Tang Jie says self-improving AI is a key step toward AGI, but must be done safely
Tang Jie, a leading Chinese computer scientist and Z.ai founder, said using AI to build even smarter systems would make AGI more achievable, but it must be done safely. He spoke at an event in Shanghai on Saturday, according to the South China Morning Post.
Chinese AI labs aim self-improvement at training infrastructure, not model weights, Leiphone survey finds
A Leiphone survey, reported by Pandaily, finds Chinese model developers pursuing recursive self-improvement by pointing AI at training pipelines, inference systems, GPU kernels, sandboxes and agent harnesses. No lab is letting AI rewrite model weights. Zhipu, DeepSeek, ModelBest, Moonshot AI, Alibaba, ByteDance and Xiaomi each disclosed a different piece of the approach.
METR says Anthropic redacted most of the evidence it found most informative in its ~100-page report
METR says it submitted a roughly 100-page full report on its assessment to Anthropic, and that Anthropic redacted a majority of the evidence METR found most informative from the public version. METR says readers must therefore rely on its judgment more than it considers ideal.
AI Running Costs Fell This Year Even as Server Prices Rose, but Data Center Curbs Test the Trend
The cost of running AI has fallen this year even as server rental prices rose. After U.S. voters elected officials who want to restrict data center expansion, The Information reports that some executives and investors worry capacity delays could stall that decline.
Comments
Loading...