47 is my crossover for serious things (e.g. Grok 4.7 is below the line and GPT 6 Sol is above the line). I mean, Opus 5.5 is way better, but GPT 6 Sol still gets the job done for anything that doesn't require design thinking.
replyAlthough I do think Luna 6 max is ok for some basic things, would never use it for coding myself.
What's the most important task you would/wouldn't trust with an agent below the line?
replyHad the same experience. I got rid of my pro sub of OpenAI. I’m really freaking impressed.
replyWhere are you inferencing DS4.1 flash reliably ?
replyI use OpenCode Go. I've also used the DeepSeek provider through OpenRouter a bit in the past and it seemed solid.
reply