I checked this website along with API pricing on OpenRouter, and this one beats Gemini 2.5 Pro (…Preview-0506 in their chart, but with a good margin so probably the non-preview too) at half Google’s API price. Nice. Admittedly their own posted benchmark, but still. If it even just competes with it, it’s a win.
Edit:
I ran my fun test on it and it unfortunately failed.
> ”How can I detect whether a user is running in a RemoteApp context using C# and .NET? That is, not a full RDP desktop session, but a published RemoteApp as if the app is running locally. The reason I’m asking is that we have an unfortunate bug in a third party library that only shows up in this scenario, and needs a specific workaround when it happens.”
It started by trying to read hallucinated environment variables that just aren’t there. Gemini 2.5 Pro had the same issue and IIRC also Claude.
The only one I have seen give the correct answer that is basically ”You can’t. There’s no official method to do this and this is intentional by Microsoft.” along with a heuristic to instead determine the root launching process which is thus far (but not guaranteed to be) RDPINIT.EXE rather than EXPLORER.EXE as in typical desktop or RDP scenarios… has been OpenAI o3. o3 also provided additional details about the underlying protocol at play here which I could confirm with external sources to be correct.
I like my query because it forces the LLM to actually reply with that you just can’t do this, there’s no ”sign” of it other than going by a completely different side-effect. They are usually too eager to try to figure out a positive reply and hallucinate in the process. Often, there _are_ these env vars to read in cases like these, but not here.
Edit:
I ran my fun test on it and it unfortunately failed.
> ”How can I detect whether a user is running in a RemoteApp context using C# and .NET? That is, not a full RDP desktop session, but a published RemoteApp as if the app is running locally. The reason I’m asking is that we have an unfortunate bug in a third party library that only shows up in this scenario, and needs a specific workaround when it happens.”
It started by trying to read hallucinated environment variables that just aren’t there. Gemini 2.5 Pro had the same issue and IIRC also Claude.
The only one I have seen give the correct answer that is basically ”You can’t. There’s no official method to do this and this is intentional by Microsoft.” along with a heuristic to instead determine the root launching process which is thus far (but not guaranteed to be) RDPINIT.EXE rather than EXPLORER.EXE as in typical desktop or RDP scenarios… has been OpenAI o3. o3 also provided additional details about the underlying protocol at play here which I could confirm with external sources to be correct.
I like my query because it forces the LLM to actually reply with that you just can’t do this, there’s no ”sign” of it other than going by a completely different side-effect. They are usually too eager to try to figure out a positive reply and hallucinate in the process. Often, there _are_ these env vars to read in cases like these, but not here.