Introducing Weave Code Max: We have hyper-optimized routing and inference so you can get $50+ in usage for just $10/month. You can use our tuned open-source models in Codex, OpenCode, Pi, or Claude Code. The most coding $10 can buy.

How we stack up

How it works

Get started:

👀

Baller

🤩🤩🤩

Heat

Looks very intriguing! I want to know more about how this works. Can I read about it somewhere? Doc?

life saver

I'm interested to try this out, but could you elaborate more on "every prompt going to the cheapest model that can handle it"? I am currently an opencode Go user but I do like to pick my own models, interested in knowing more about your offerings

Our early users prefer the router to pick the right model as it's been exhausting keeping up with every new release and burning tokens testing what each model is good/bad at. Give it a shot and lmk your thoughts.

C'est une offre incroyable pour le prix.

nah, only if the spend stays capped. i hit the same “cheap” route, the retry loop burned the budget, and i rolled back to the old model.

Capped unless you add credits

The best coding agent on the market

By routing each request to tuned open-source models and optimizing inference, Weave aims to stretch a fixed subscription budget further.

We did shut downsizing dev (doubling the subscription) because - providers fight routers, for example, Claude signs commands with its private keys, so you can’t effectively route between providers back and forth - Quality and stability matter much more than cost savings to the customers we’ve spoken to - When you route, you break the cache, which makes it financially unfeasible in many cases But maybe you guys will find a way through! @daltonc

$50 in usage for $10 is wild if the models actually hold up under real workloads. The tuned open-source angle is smart, but I'd trust it more once I see how the smaller models handle edge-case bugs.

Give it a test. Think you’ll be impressed

Will do. Curious if the smaller models handle complex refactoring or just simple completions, that's where most tuned models fall apart.

长任务跑下来稳不稳啊？

ok but can i say when to use lesser models somehow? For example there are use cases (that it might not understand) that can be easily done by smaller/cheaper models

the $10 price point is kind of absurd for that much coding headroom

@daltonc Where are the open source models hosted? US?

@daltonc All hosted in US

@daltonc awesome thanks!

So wait, we don't know which models are being served under the hood, or even which ones are a part of the underlying stack? Even the docs page seems to be missing. Will the analytics give an idea here? No offence but this seems to be having a lot of trust me bro energy