So you want to use OpenRouter?

Thanks for the insight here.

One thing to note about the first graph: nobody is doing as well as the first part on tool calling, and it's not close.

This might be the fault of the other providers, but it's probably just something slightly different that the first party does with the model inference program than anybody else, and that's not sure to weights it's due to vLLM twiddling (or whatever) and probably becuase the first part actually uses their own customized inference program rather than the standard methods that all the third party providers use. This isn't nefarious, it's just the challenge of these sorts of stochastic systems.

Having been in science for decades now, and seen benchmarking across many different fields, these results are completely expected for me. LLM serving is not mechanical, it's hard to get right and has lots of unknown footguns. Even something as extreme as scrambling a matrix will still likely get results that are nearly as good as normal, and if there's a bug deep in vLLM or the tensors metadata that results in that, then it's going to be pretty hard to find unless you're an active researcher with knowledge of the particular model you're running inference on. I kind of doubt that's happening here, but maybe!

In the scientific literature, when benchmarking methods, everybody's own method performs best in their own hands. Some attribute it to researchers gaming benchmarking for publication purposes, but I think it's just what we see here: the people who made a method are just the best at using it because they know all the quirks and use it best.

Programmers are not used to thinking with that nuance, and jump to conclusions about lying about quantizations, etc., but this is really just an unavoidable part of AI/ML methods: when things aren't perfect they're still pretty good and it's going to take the model creator to truly debug it. At least until the open weights ecosystem gets a lot better at ensuring reproducibility, and model cards are nowhere detailed enough for that to happen yet.

I also believe some providers fallback to another model entirely. I was recently using Kimi K3 and saw that some requests had no reasoning trace whatsoever. Unsurprisingly, those requests were routed to the less reputable providers (Sail Research).

Yes to all this but more. The thing that made me leave and go to a single provider was token caching. I have to keep blocking providers that don't properly cache. I see performance tank and then I look in the logs and a new provider has been rotated in and every call to them is uncached because they are clearly broken. This has happened a few times now and essentially destroys cost savings (these providers also often have terrible quality). Don't they monitor for simple things like this? Their own logs show how clear this pattern is for some providers. Simple cache % stats would allow them to block providers nearly instantly.

Until yesterday I used openrouter mostly as a portal to the big providers: google, anthropic, openai, without having to maintain keys and accounts and credits for all of them separately.

Yesterday I wanted to do a bit of benchmarking a prompt across multiple models. Small requests. Outside the big providers, the experience became awful. This explains that experience.

[flagged]