BTW, immediately after posting the message above, I saw that GPT 5.6 became available on Openrouter - I hadn't even seen in the news that it had been released to the public yet. I keep my eyes on https://openrouter.ai/models every day, to see which new models are available. It's better than any news source, to keep up with all the new models being releases by companies around the world. Instead of reading hype about all the models, you can simply try them for yourself.

So I popped openai/gpt-5.6-sol into the Jan chat harness (get Jan for free at https://www.jan.ai) and tried the prompt linked below (that prompt has become a favorite of mine for getting a quick sense about how much knowledge a model has about obscure topics). Sol did a great job:

https://com-pute.com/nick/sol_first_prompt.txt

And to be clear, I just like testing new models quickly in the Jan.ai chat interface, because it only takes a few seconds to set up a new model in Jan (Jan runs locally on your PC and connects automatically to models provided by OpenRouter and other LLM API providers). But I could have done the exact same thing in Pi. You can chat in Pi - you don't necessarily need to use it to work on agentic tasks, or to develop software. Jan is a quick and convenient point & click chat tool that I like to use it for short conversation interactions.

Notice that Sol is expensive. The response linked above cost 6 cents to generate. A response from Deepseek V4 Flash would typically cost 150-200x less than one from Sol - so I tested the exact same prompt with it and it actually cost even less than that: $.000247 (247x less than Sol). Both responses are included in the link above.

BTW, immediately after posting the message above, I saw that GPT 5.6 became available on Openrouter - I hadn't even seen in the news that it had been released to the public yet. I keep my eyes on https://openrouter.ai/models every day, to see which new models are available. It's better than any news source, for keeping up with new model releases.

So I popped openai/gpt-5.6-sol into the Jan chat harness (get Jan for free at https://www.jan.ai) and tried the prompt linked below - that prompt has become a favorite of mine for getting a sense about home much knowledge a model has about obscure topics. Sol did a great job:

https://com-pute.com/nick/sol_first_prompt.txt

And to be clear, I just like testing new models quickly in the Jan.ai chat interface, because it only takes a few seconds to set up a new model in Jan (Jan runs locally on your PC and connects to models provided by Openrouter and other LLM API providers). But I could have done the exact same thing in Pi. You can chat in Pi - you don't necessarily need to use it to work on agentic tasks, or to develop software. Jan is just a quick point and click tool, so I use it for quick little interactions.

Notice that Sol is expensive. The response linked above cost 6 cents to generate. A response from Deepseek V4 Flash would typically cost 150-200x less than one from Sol - so I tested the exact same prompt with with it and it actually cost even less than that: $.000247 (both responses are included in the link above).