Relace

This account is verified. Learn more

@relace_ai

Models and infra for AI code generation.

relace.ai Joined September 2025

9 Following 773 Followers 47 Posts

Latest posts 5

Relace @relace_ai Nov 5

We're now live on the @concentrateai gateway!

Relace has spent the past four months aggressively driving down the cost of every open weight model we serve. Concentrate charges zero fees on top, so that price is what you pay.

Glad to partner with them to continue to make intelligence more accessible.

Relace @relace_ai Nov 4

It took us 401 minutes to go live with Kimi K3.1 Pro from the moment the weights went public.

Next time, we'll beat that record.

Relace @relace_ai Nov 2

We're expanding our presence on the pareto frontier with both the fastest endpoint and the cheapest!

Eitan Borgnia Verified account @EBorgnia Nov 2

Over the weekend, we released our fast endpoint for DSv4.1 Flash.

Relace is currently #1 on OpenRouter, and there's still a lot of room for improvement!

View quoted post on X

Relace @relace_ai Sep 24

If you want to spend all day working on inference optimization, we're hiring. relace.ai/about-us

relace.ai About Relace - Building AI Tools & Models for Smarter Coding Learn how Relace was founded to solve core challenges in AI-driven coding. Discover our mission, team, and vision for empowering developers with specialized models and infrastructure.

Eitan Borgnia Verified account @EBorgnia Sep 24

Relace is now also the cheapest provider of DeepSeek v4.1 Flash on OpenRouter.

Agent traces are 96%+ input tokens, so GPU utilization is dominated by prefill time. However, most of those tokens are not relevant for the next response.

The encoder/decoder split allows you to use a lossier, cheap submodel for prefill without affecting the performance of decode (i.e. quality of responses).

It also means you need to shuttle KV between GPUs a lot, which is why they redesigned attention (again) and made the KV 4x smaller.

Fully taking advantage of the theoretical gains is a deep engineering problem that DeepSeek left as an exercise for the reader.

View quoted post on X

Relace @relace_ai Sep 11

Today we did 37% of Deepseek v4 Flash, 25% of GLM 5.3 Flash, and 1.6% of Kimi K3 traffic.

Eitan Borgnia Verified account @EBorgnia Sep 11

Relace just hit 1T tok/day on OpenRouter

View quoted post on X