Pi v0.80.6 ships max thinking, tiered pricing
Pi v0.80.6 introduces a 'max' thinking level for advanced reasoning on GPT-5.6 and adaptive Claude models, alongside request-wide input-token pricing tiers for precise long-context cost tracking. The release also includes fixes for post-compaction token budgeting, Anthropic message conversion, and home directory path expansion.
Granular reasoning controls and tiered token pricing are becoming table stakes for AI coding agents as context windows and model capabilities balloon.
- –The new 'max' thinking level allows developers to choose when to spend more time (and tokens) on reasoning with models like GPT-5.6 and adaptive Claude variants.
- –Request-wide pricing tiers reflect the non-linear pricing models of modern LLM APIs, ensuring developers have accurate cost estimates for long-context runs.
- –Crucial bug fixes for post-compaction budgeting and Anthropic message conversion ensure the agent loop doesn't fail during extended, multi-turn programming sessions.
DISCOVERED
2h ago
2026-07-09
PUBLISHED
2h ago
2026-07-09
RELEVANCE
AUTHOR
PiChangelog