REAX

Decisions at reflex speed: fast open decision models.

[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

🧵

[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 👇

[REAX @reaxlabs]

@AntSeed Excellent, thank you!

[Antseed @AntSeed]

Next router on Antseed: Decision Models 🐜

An auto router that picks the model and seller for every request, with an open-source implementation you can inspect and adapt to your own needs

[REAX @reaxlabs]

@decisionmodels_ @AntSeed Powered by Bittensor subnet 92, REAX!

[Decision Models @decisionmodels_]

Decision Models is now on @antseed 🎉

• Our decision models, including an EU tier
• Our Auto Router: it picks the best model for each request to give you excellent price/capability

[REAX @reaxlabs]

@antoine_chaffin Kudos for your Decider models, Antoine! Excellent work

[Florian S @airesearch12]

Nice, one of our new organic customers of the EU Tier of https://t.co/aR2C6rJeeV needed his rate limit to be increased to 1,200 requests per minute. ☺️

[Mark Jeffrey @markjeffrey]

Bittensorians, please take note.

I have informed him of @decisionmodels_ in answer to #2.

[REAX @reaxlabs]

Correction on "starting right now":

From now on you can prepare with the code, build your engine against the s1-fast reference and test it locally.

The competition itself opens tomorrow, Tue 20 Oct, 18:00 UTC.

That's when submissions start counting.

[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.

[REAX @reaxlabs]

@airesearch12 * oh and saved money means more margin

[REAX @reaxlabs]

@decisionmodels_ YES

[REAX @reaxlabs]

Code here:

https://t.co/DkipOWgqVl

[REAX @reaxlabs]

Thanks for calling us a highlight 🫡

[Andy ττ @bittingthembits]

🚨 One week. 8 $TAO subnets. 8 different industries. Zero of them sell the same thing.

Oct 12 to 18, from @opentensor 's weekly highlights:

• SN92 @reaxlabs: open inference engine up to 4x faster on the same GPU, plus an MIT-licensed open-weights router (26 ms on a CPU)
• SN65 @TPN_Labs: Ministral-3-8B made 40% smaller, full 32K context kept
• SN95 @actualinc: 10,500 tok/s on a single NVIDIA DGX Station, launched with NVIDIA
• SN53 @engyai: 50B tokens in a day, then nearly 102B the next day on its public dashboard
• SN25 @urnetwork: 100,000+ existing providers moving onto mainnet as miners
• SN44 @webuildscore: computer vision now at a Shell motorway station, after the Avia rollout
• SN80 @openroboto: robot training data devices with on-chain identities on Solana
• SN28 @say_gm_: Claude, GPT and Gemini in one desktop app, priced below list

Decision models. Model compression. Inference hardware. Token serving. Bandwidth. Vision. Robotics data. A consumer app.

That's the whole AI stack, shipped in 7 days.

To match this week elsewhere you'd need:

• 8 companies
• 8 cap tables
• 8 budgets.

Big Tech does it with a payroll.

Bittensor does it with a market.

One token pays all 8 teams to compete, and the market decides who keeps getting paid.

Name one project in crypto that ships across the stack like this.

I'll wait.

Yes, true, its not perfect and most benchmarks are team-reported. Most Subnets are early and uneven.

But a week like this, every week, is how a market gets mispriced.

Don't trust me.

Check it yourself: https://t.co/sKGaaOa9HJ, https://t.co/d6C5vMUh5u, and Actual's NVIDIA case study.

$TAO / DYOR.

[YVR τrader @YVR_Trader]

Decision models may be bigger than chatbots.

ChatGPT writes. Decision Models decide.

Every app will need millions of tiny, instant choices: route a ticket, flag content, pick the right model, approve a workflow, classify an image.

Using a frontier LLM for that is like calling a supercomputer to flip a light switch.

@reaxlabs #SN92 on $TAO is building the infrastructure for this new layer.

•Fast, structured decision models that can be 100–200x faster and cheaper than general LLMs in the right use cases
•A live commercial API with paying customers found before the public launch
•Miners competing to serve inference faster and train specialized decision models
•A router layer that sends each request to the cheapest model capable of handling it
•Real-time potential in agents, robotics, moderation, routing and enterprise software

This is not another chatbot subnet.
It is a decentralized factory for decision intelligence: many small expert models, continuously improved by an open miner network, served through one product.

Frontier AI is a capital war. Decision models are different: small enough for independent builders to compete, valuable enough to sit inside every piece of software.

If AI becomes the nervous system of software, @decisionmodels_ wants to be the network that makes those decisions.

This brainchild of @airesearch12 could be a juggernaut in the making.

[REAX @reaxlabs]

@TheNDubbs @bittensor Yes!!! Thanks, we try our best

[REAX @reaxlabs]

@harnyx_ai @tao_Alph @RespanAI @NaceAI @StepFun_ai @decisionmodels_ Hey yeah, feel free to contact us, ideally via @airesearch12

[REAX @reaxlabs]

@decisionmodels_ 🔥

[Decision Models @decisionmodels_]

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

[REAX @reaxlabs]

@decisionmodels_ 🔥

[REAX @reaxlabs]

Quick moment to step back, because this was a big week

• 3 days ago REAX went live. Today 217 miner slots serve decision models and 65 independent miner wallets are earning... almost all traffic routed to the network is now answered by your GPUs, with zero failed verification checks.

• The first model-training competition has its first two champions
=> demand forecasting (UID 71, 91.1 vs the 38.2 baseline) and
=> risk assessment (UID 251, 64.2 vs 31.7).

Both more than double what we started with.

Both passed a confirmation draw, a head-to-head playoff, and copy/twin checks against every earlier entry.
Rewards start with the next weight sets.

• Many fixes this week came straight from the mining community:
real-speed probes per hotkey, keep-alive for qualified entries, re-registration without copy flags, a fairer small-GPU tier, and Senpaii's confirmation-margin PR.
Next: protection for queued entries, several machines per hotkey, and less registration churn.

Thank you all, especially dear miners, for the hard work, the bug reports and the patience.

A young subnet has rough edges, and you're the ones helping us smooth them out.

Great things lie ahead! 🫡