i’m giving openrouter a try now, has most of the models so good for comparing, but token based pricing, curious if that ends up becoming too pricy… although maybe not with a harness like pi! claude code will send 50k tokens in tools for a simple ‘ping’ o.O
self hosted 30B qwen is actually suprisingly capable as well! and glm-4.7 flash (q4_k_m) fits on my gpu and is great (but tends to get into infinite loops :( )
i’m giving openrouter a try now, has most of the models so good for comparing, but token based pricing, curious if that ends up becoming too pricy… although maybe not with a harness like pi! claude code will send 50k tokens in tools for a simple ‘ping’ o.O
self hosted 30B qwen is actually suprisingly capable as well! and glm-4.7 flash (q4_k_m) fits on my gpu and is great (but tends to get into infinite loops :( )