A wild DeepSeek has appeared 👀 chutes.ai/app/chute/14a9… huggingface.co/deepseek-ai/De…

Working on a hotkey signed TLS cert helper thing, pretty neat, simple. Create a private key and cert, include a signature with your hotkey of the CSR (roughly). Client library to connect, get the cert, check the signature against the ss58, store the cert and use it for strict

Decentralized compute is winning. We don't have one datacenter, we have dozens. We don't have one SRE team, we have nearly 100. Latest example: DeepSeek-R1-0528. 100% uptime, day zero support, 4x more tokens on openrouter than all other providers combined (and go check the

FYI the stake transfer was to an AI ninja to help with a few things 🪂

TAO.app Savant is now powered by SN4 Targon Manifold and SN64 Chutes 🚀🤝 Our new system dynamically routes between DeepSeek and Claude-4-Sonnet to deliver the best user experience. We’re evolving fast and value your feedback! Try it now at

It brings me great joy to see const be able to fulfill the vision of dtao, further decentralize bittensor/OTF, and join the grind with Affine. What a legend. I don't know if he knows what he got himself into though 😅

Chutes Payments with fiat are now Live in Beta 🪂 With our next step towards mass revenue, we are accepting fiat payments for all paid models. You may consume your balance through both the playground apps and our API. All Revenue is auto-staked to the Chutes Alpha Token.

Thanks for having me Mark Jeffrey!

We have just launched a new free tier for Chutes which is currently a 200 request global limit for all models. Beyond this, models are now all paid unless otherwise specified. (Re-roll prompts have special, separate rate-limits at 1/10th of an invocation towards the quota)

Unfortunately a few bad apples are causing chaos. For example, via pattern analysis and request fingerprinting we can assess probability of a user being the same as another, and some entities are badly abusing the free tier limits via multiple accounts.

Very well said.

R1T2 Chimera is available on Chutes Thanks to Jon Durbin and team for making it available within 24 hours!

The changelog for the next chutes update is going to be approximately 8 trillion pages.

It seems, according to deploy docs, the minimum viable configuration is 16 h200s. Unfortunately we don't support tying nodes together (yet), but b200/mi300x support is just around the corner. I think this is the perfect use case for our first deployments of those GPUs.

Actually it does work fine on 8xh200 with 65536 ctx (and 3 max concurrent requests if you want to allow the full ctx for each). chutes.ai/app/chute/35cf… We'll still move it over to the larger GPUs when we can to support the full context and higher concurrency!

This SGLang bug is the reason Kimi-K2-Instruct is so unstable/crashing. I've already tried disabling tools/structured output/etc. which uses logit bias. Something about the batch size changing mid-flight here maybe?