Xiaomi debuts open-weight omnimodal models MiMo-V2.6 Pro and Flash; Pro allegedly performs “on par with Opus 5 and GPT-5.6 Sol across most agent benchmarks”
Frontier intelligence, all the modalities, built in public. Today, we are releasing and open-sourcing the MiMo-V2.6 series.
Xiaomi
Context & Ripple Effects
Xiaomi’s MiMo program has progressed from the open-weight MiMo-V2-Flash in December 2025 to a 1T-parameter Pro model in March 2026, whose performance Xiaomi positioned near leading proprietary systems. In April, Xiaomi made MiMo-V2.5 and V2.5-Pro available under the MIT License, establishing open distribution as a recurring part of the line’s strategy.
MiMo-V2.6 extends that approach from agentic tasks into an omnimodal Pro-and-Flash release. Xiaomi’s claimed parity with Opus 5 and GPT-5.6 Sol is the commercial test: an open-weight alternative becomes more consequential when buyers can assess it against proprietary frontier options.
First-order effects
Developers can deploy, inspect and adapt Xiaomi’s open-weight MiMo-V2.6 Pro and Flash models rather than treating omnimodal and agentic capability solely as an API purchase.
Xiaomi’s parity claims put MiMo-V2.6 directly in evaluation sets alongside Opus 5 and GPT-5.6 Sol, even though those benchmark comparisons remain Xiaomi’s assertions.
Second-order effects
Open-weight model vendors face pressure to compete not just on downloadable weights but on multimodal and agent-oriented performance, the areas Xiaomi is using to position MiMo-V2.6.
Buyers evaluating proprietary models gain a portable alternative to test against hosted offerings, increasing the importance of deployment cost, tooling and support alongside raw benchmark scores.
Third-order effects
If independently reproducible evaluations support Xiaomi’s claims, frontier-model competition shifts further toward services, infrastructure and integration rather than exclusive control of model weights.
Repeated MIT-licensed MiMo releases point to an open-weights market where model portability can reduce switching costs and weaken a provider’s ability to rely on model access alone as a moat.
The trend: Open-weight AI is moving from narrower reasoning releases toward multimodal, agent-capable models positioned as substitutes for proprietary frontier access.
“We're open-sourcing Pro and Flash, MiMo-V2.6-Distill-Qwen-9B, the technical report, 7K+ RL task environments, an end-to-end RL framework and composable mini-harnesses.” wow 🔥
How?!! Xiaomi has just released their new OPEN SOURCE model MiMo-V2.6-Pro: - multimodal text/image/audio/video - score ~ same as GPT-5.6-Sol Max - 9x cheaper input tokens - 23x cheaper output tokens 💀 Also about 20x cheaper than Opus 5... and the weights are already on Hugging Fa…
The difference between the frontier of closed source and open source has been reduced by a good margin AND this time it is a “not” a TOP-3 Chinese labs.
Alert: Xiaomi just open-sourced MiMo-V2.6. Two natively omnimodal models, Pro and Flash, under MIT license 🔥. Pro scores 46.32 on the Artificial Analysis Intelligence Index, the highest of any open model to date. weights: https://huggingface.co/...
the most insane part, they will release ~7k RL training data and the framework leading to this top 6 model on AA, they also shipped the model + tech report less than 1 week after starting the final RL run pushing both intelligence and openness level, huge congrats
Xiaomi is contributing a LOT to open research with this release. All of their training run details are basically out there in the open — reward plots, hyperparameters, data mixtures, even costs. They also release a subset of their RL envs and a smaller model for people to play ar…
Introducing Xiaomi MiMo-V2.6 — Pro & Flash. Frontier intelligence, all the modalities, built in public. 🔹 Two omnimodal models, advancing through scaled reinforcement learning 🔹 Pro performs on par with Claude Opus 5 and GPT-5.6 Sol across most agent benchmarks 🔹 Pro scores 46 on…
MiMo-V2.6-Pro debuts as the top open weights model on the Artificial Analysis Intelligence Index (46). At $0.13 per Intelligence Index task, it lands on the Intelligence vs. Cost per Task Pareto frontier @Xiaomi has just released MiMo-V2.6-Pro, an open weights model with major ad…
Update every prior on post-training scaling laws: They hit SoTA with ~45k RL tasks, nearly all synthetic, with $2.6m of Chinese compute (~$1m in US) That's a 2023 era pre-seed! Human data is stifling the US frontier, and neolabs are prematurely scaling compute.
Xiaomi just dropped Mimo 2.6 Pro which claims to be the best open source model. I played with it, and was pretty impressed: - Insanely cheap: With some standard assumptions, its 15x cheaper than Kimi K3, 6x cheaper than GLM 5.3 and 2x cheaper than DeepSeek V4 and only 2x more exp…