Decision Models

Fast, low-cost open decision models on one Jev-compatible API. Route, score & approve with probabilities.
EU-hosted option.
Powered by @reaxlabs (SN92).

[Decision Models @decisionmodels_]

@reaxlabs faster => happier customers 💪

[Decision Models @decisionmodels_]

The engine inside of https://t.co/1P2thBTuzW, and the reason why we most likely can serve these models faster than you may have seen them anywhere else.

[REAX @reaxlabs]

The REAX engine serves Perplexity Decider v1.1 up to six (!) times faster than its reference server.

Free for individuals and small companies, even commercially. Code in next post.

Performance improvement competition opens right now: Beat the s1-fast reference, win the track.

[Decision Models @decisionmodels_]

@airesearch12 💪

[Decision Models @decisionmodels_]

Use leading Jev class models
- blazing fast and
- for crazy low prices
on https://t.co/QIARRjBNRC

[Decision Models @decisionmodels_]

@reaxlabs 🔥

[Decision Models @decisionmodels_]

@reaxlabs 🔥

[Decision Models @decisionmodels_]

@airesearch12 💪

[REAX @reaxlabs]

Introducing REAX, Bittensor subnet 92!

An open network about AI decision models:
- miners host, train and spped-up models
- validators test correctness
- and customers use them through https://t.co/2H3q6dIK9j

A gaming GPU is enough to join. https://t.co/rPV6jcduIv

🧵

[Decision Models @decisionmodels_]

How it compares on our own held-out routing test (2,175 synthetic requests in 22 languages, labels from a 5-LLM consensus, method in the model card):

It picks the same model as the gold label 74% of the time, the best of the decision models we tested, at 26 ms on a CPU.

On JevBench:
As its operator, we never rank our own models there. And a specialist like this isn't a fit for general decision-model benchmarks anyway, so we measured it on the task it was built for.

[Decision Models @decisionmodels_]

We're releasing s1-llm-auto-router as open weights:

A small decision model that reads a request and picks the right LLM for it.

▸ 26 ms on a CPU, 22 languages, MIT license
▸ Our first use-case-specific decision model, from https://t.co/VklNqP0jeV and Bittensor Subnet 92 (REAX)
▸ The best of the decision models we tested for this use case.
▸Now live on https://t.co/VklNqP0jeV.

We're also announcing a joint project with msg group:

msg will evaluate the model in an internal project to make its use of LLMs more efficient. msg is one of Germany's top 5 IT consultancies (€1.5B revenue, 10,000+ employees).

Thank you, Dr. Ralf S. Engelschall (@engelschall), CTO of msg and one of the founding members of the Apache Software Foundation.

We believe use-case-specific decision models will become a huge market. More from us soon.

https://t.co/6UE8FaVQwW

[Decision Models @decisionmodels_]

Served by REAX, an open inference network on Bittensor subnet 92.
A gaming GPU is enough to participate.
https://t.co/VxgDKJ89dn

[Decision Models @decisionmodels_]

Hi, we're DecisionModels.

The best open decision models, served fast through one API:

→ s1-pro runs vLLM-SR d3, the most capable decision model in the world *
→ up to 4x faster on our own REAX engine
→ s1-fast from $0.019 per 1M input tokens (about 1.6x cheaper than Jev), output free
→ EU-hosted option with a DPA, built for GDPR and the EU AI Act

https://t.co/VklNqP0jeV

* according to both the Decision Index on Hugging Face and the Capability Score on JevBench (our founder's own project)

[Decision Models @decisionmodels_]

@rudrakshkarpe @airesearch12 thanks

[Decision Models @decisionmodels_]

@michael_chomsky @airesearch12 that's the spirit 💪

[Florian S @airesearch12]

Introducing DecisionModels @decisionmodels_

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:

https://t.co/uLqryOaB5Y

➡️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 👇