Introducing DecisionModels Decision Models I've spent the last four weeks benchmarking decision models (I run JevBench). Today I'm launching the place to actually use the best ones: ➡️Top open models ➡️Up to 4x faster on our own REAX engine ➡️From $0.019 per 1M input tokens. Expect more reveals soon,... and read 👇show more

DecisionModels is powered by REAX, Bittensor subnet 92. ➡️A place where anyone with a GPU can host these models. ➡️Where the community of AI engineers gets incentivized to work on REAX engine, our high-speed inference engine for decision models (and for upcoming other work on decision models) Visit to get started, quickstart below.

Disclosure, since I run the JevBench benchmark (that is used by companies like Microsoft when launching their decision models) My work on REAX and DecisionModels will not influence this independent and fair evaluation. Methods and raw results are public at Check them.

Thank you @const_reborn for funding the subnet, and the @chutes_ai team, where I learned how great Bittensor subnets work and what decentralized inference can do. The best models should be easy to use, fast, and served by the people, for the people.

@decisionmodels_ Germany taking over. lfg

@decisionmodels_ Hahaha, oh no, please not

@decisionmodels_ Congrats Florian on the launch of @decisionmodels_

@decisionmodels_ Thanks You!!!

@decisionmodels_ so insanely cool!

@decisionmodels_ Thanks Laurits!!! Do you have any ideas for that service? let me know

@decisionmodels_ typesafe team has been very vocal in preferring the nomenclature of system one models, what made you decide to go with decision models as the name

@decisionmodels_ Congrats on the launch :)

@decisionmodels_ Thanks Nick! :)

@decisionmodels_ the price to performance ratio looks so good!

@decisionmodels_ Thanks Aman! That's exactly what we optimized for 🙏

@decisionmodels_ interesting, congrats!

@decisionmodels_ Thanks! 🙏

@decisionmodels_ Congrats, checking it out!

@decisionmodels_ Thanks Jason! 🙏

@decisionmodels_ Now Bittensor has the ~decision wars~, aka the jev wars. Who will win @zils_ai or @decisionmodels_?

@decisionmodels_ @zils_ai Haha, may the fastest model win 😄

@sebyrubino @decisionmodels_ @zils_ai Or most accurate 😉

@airesearch12 @decisionmodels_ @zils_ai And we are off to the races! 🏇🏇

@decisionmodels_ Congratulations!

@decisionmodels_ Thank you so much. Let me know what I should implement

@decisionmodels_ Congrats, this looks awesome!

@decisionmodels_ All the best with this project! Florian is awesome.

@decisionmodels_ Thank you Omar, that's very kind! 🙏

@decisionmodels_ Congratulations on finally getting this on Bittensor. I know it's going to turn out excellently.

@decisionmodels_ Thank you! 🙏 Excited to build this on Bittensor.

@decisionmodels_ Legend

@decisionmodels_ Danke dir, Marko! 😄

@decisionmodels_ Nice work !

@decisionmodels_ Congrats! Und das aus Deutschland 💪

@decisionmodels_ Congrats on the launch How does your REAX engine achieve that speed boost

@decisionmodels_ Thanks! Decisions need almost no generation, so the REAX Engine is built prefill-only, with quantisation (FP8/NVFP4) and batching tuned for short decision outputs. It's open source:

@decisionmodels_ four weeks of benchmarking before shipping the place to use them

@decisionmodels_ Congratulations and more growth to you 🫡

@decisionmodels_ Thank you!!!

@decisionmodels_ did you notice a strong correlation between backbone frozen model parameter size and performance? would be curious to see how adding parameter to the frozen model increases performance

@decisionmodels_ Congrats Florian. This looks awesome. How are you getting GPUs?

@decisionmodels_ Thanks Mohit! Dedicated RTX PRO 6000s in EU data centres for now, and over time more and more capacity from REAX miners on Bittensor subnet 92.

@decisionmodels_ Congrats Florian! 💪

@decisionmodels_ Thanks Humberto! 🙏

@decisionmodels_ Congrats on launch!

@decisionmodels_ Thanks man! 🙏

@decisionmodels_ Congrats to the launch! Florian

@decisionmodels_ Thank you Xunzhuo! And thanks for vLLM-SR, d3 is what powers our s1-pro 🙏

@decisionmodels_ 🔥🔥🔥

@decisionmodels_ 🙏🚀

@decisionmodels_ Excited to watch this grow into a blue chip subnet! This is fantastic brotha

@decisionmodels_ Thanks brotha! 🙏

@decisionmodels_ Congrats on the launch! 🤩 Check out Agentic Wear, if you find time for it! It's my first ever public app launch!

@decisionmodels_ Thanks, and congrats on your first launch too! 🚀

@decisionmodels_ Congrats bro .. You have been cooking .. More power to you.

@decisionmodels_ Thanks bro, means a lot! 🙏

@decisionmodels_ anytime bro

@decisionmodels_ So excited to see this project. ❤️

@decisionmodels_ Thanks! 🙏

@decisionmodels_ Congrats

@decisionmodels_ Thank you Volodymyr! 🙏

@decisionmodels_ JEV to the Moon!

@decisionmodels_ Thanks

@decisionmodels_ Been following you since Jev Bench, awesome work

@decisionmodels_ Thank you, that means a lot! 🙏

@decisionmodels_ Big you up bro! Looking forward to trying it out

@decisionmodels_ Thanks bro! Let me know how it goes 🙏

@decisionmodels_ 跑分靠前的，实际做决策也更稳吗？