Pi supports 30+ providers with mid-session model switching, so you can match the model to the task. Its minimal tool surface keeps prompts lean, which lets mid-priced models punch above their weight. Claude Sonnet 4.6 is the quality default; for value, DeepSeek V4 Pro, Kimi K2.7 Code, and GLM 5.2 deliver near-frontier coding at open-weight prices.
The quality default — strong multi-file edits that Pi's lean context makes the most of.
The value champion — near-frontier quality at a fraction of the cost, open weights included.
Coding-tuned with great tool use — a strong cheap default for Pi's read/write/edit/bash loop.
A capable open-weight flagship with a 1M context for long sessions between compactions.
A well-priced GPT option for those who prefer the OpenAI family.
The model picks the moves; the agent runs the loop, the tools, and the guardrails. Once you've chosen a model, see which agent gets the most out of it.
Every model on this page is included with Standard Compute
