I briefly tried some Chinese models with pi.dev. It was fine, I think I mostly preferred the UX of pi as a harness to Claude Code, but my experience with the chinese models was worse than with Claude's models. They just felt significantly dumber than Sonnet 5.
Maybe that's partially up to the harness though, I'm not sure.
I briefly tried some Chinese models with pi.dev. It was fine, I think I mostly preferred the UX of pi as a harness to Claude Code, but my experience with the chinese models was worse than with Claude's models. They just felt significantly dumber than Sonnet 5.
Maybe that's partially up to the harness though, I'm not sure.
I've found DeepSeek v4 models to run well in pi.dev
I agree they're not as capable as Claude Sonnet, but they're much cheaper.