Builders are arguing today about Xiaomi's trillion-parameter MiMo-V2.6-Pro-RL landing on Fireworks — and specifically whether its claimed 46 score on the AA Intelligence Index survives contact with real workloads. The verified fact underneath: the deployment is live.
Key facts
- Fireworks' model library lists MiMo-V2.6-Pro-RL as Ready — on-demand (dedicated-GPU) deployment, created September 25, 2026, 1.02T sparse-MoE parameters, 1040k context, linking the XiaomiMiMo/MiMo-V2.6-Pro-RL Hugging Face repo.
- Fireworks is not the first to serve it — OpenRouter already routes MiMo-V2.6-Pro via GMICloud, DeepInfra and Xiaomi itself.
- The 46 AA Intelligence Index score is a vendor-cited claim — Fireworks presents it as the highest of any open-weight model; we have not independently verified it, and X discourse is already debating whether the model is 'benchmaxxed'.
- The model itself shipped earlier — DeAI covered the open-weights release with MIT-licensed RL training code on September 22; this is a distribution event, not a release event.
What's driving the conversation
The announcement post from @FireworksAI_HQ is the velocity driver — the only high-velocity item in today's X sweep. Provider-availability announcements for the current open-weight favorite move fast, and this one carried the usual two arguments. One camp treats a trillion-parameter open-weight model reaching a major commercial serving provider as validation that open weights are now table stakes for serious inference infrastructure. The other camp, including a Reddit thread on the launch, calls the model 'benchmaxxed' — arguing the 46 AA Intelligence Index score reflects benchmark-tuned capabilities that will not generalize to real workloads. Both positions are commentary; neither is yet evidence.
The substance
What is verified: the Fireworks model page exists, lists state Ready, creation date 9/25/2026, 1.02T sparse-MoE parameters, 1040k context, on-demand (not serverless) deployment, and links the XiaomiMiMo/MiMo-V2.6-Pro-RL Hugging Face repo. What is claimed: the 46 AA Intelligence Index score, the "highest-scoring open-weight model" framing, 42B active parameters, multimodal quality, and any endpoint-performance figures. None of those claims have been independently checked; they are provider marketing until someone replicates them.
Why builders are watching
The actionable question is the one our Fireworks-alternatives and Together-vs-Fireworks coverage already asks: which providers serve the biggest open weights, on what terms — on-demand versus serverless, fine-tuning support, rate limits — and what the benchmark claims are worth. This is also the second provenance-adjacent question on this model in a week, following the provenance questions DeAI covered September 23. Watch how quickly independent evals of MiMo-V2.6-Pro-RL land, and whether the on-demand listing expands to serverless.
Questions
- What actually launched on Fireworks?
- Fireworks AI's model library now lists MiMo-V2.6-Pro-RL — Xiaomi's trillion-parameter sparse mixture-of-experts model — as Ready, created September 25, 2026, for on-demand (dedicated-GPU) deployment, with 1040k context and a link to the XiaomiMiMo/MiMo-V2.6-Pro-RL Hugging Face repo. Fireworks is joining GMICloud, DeepInfra and Xiaomi itself as a serving provider.
- Is Fireworks the first to serve MiMo-V2.6-Pro?
- No. OpenRouter already routes MiMo-V2.6-Pro through GMICloud, DeepInfra and Xiaomi itself. Fireworks' addition is a distribution event, not a release event — the model's open-weights drop was covered September 22.
- What is the AA Intelligence Index score, and is it verified?
- Fireworks cites 46 on the Artificial Analysis Intelligence Index as the highest for any open-weight model. That is a vendor-cited third-party benchmark claim DeAI has not independently verified; a Reddit thread on the launch already uses the word 'benchmaxxed,' and the number should be treated as marketing until replicated.
- What is disputed about MiMo-V2.6-Pro-RL?
- The benchmark claims. Fireworks presents the model as the top open-weight scorer on the AA Intelligence Index at 46, with multimodal quality and endpoint-performance claims on top. None of those figures have been independently checked this run; the verified fact is the deployment's existence, its 1.02T-parameter size, its 1040k context, and its September 25 creation date on Fireworks' library.
- Why does provider availability matter for builders?
- The same open weights can be served by many providers on very different terms — on-demand dedicated GPUs versus serverless endpoints, different fine-tuning support, rate limits and pricing. Which providers serve the biggest open models, and on what terms, is the actionable question the Fireworks and Together comparisons track.
Sources
About DeAI
DeAI is an independent publication covering open-weight AI models, private inference, and decentralized infrastructure — the tools for running AI you actually control. We test providers on price, privacy, and refusal behavior and publish the numbers, not the vibes. DeAI is powered by Morpheus (mor.org), a decentralized inference marketplace, and covers it on the same terms as every other provider.
Powered by Morpheus and StrandCMS
Morpheus is a decentralized inference marketplace, covered on the same terms as every other provider — we rank it wherever the data lands. StrandCMS is the open-source, agent-first framework this site is built on.