release aidev SIG 3/5

Pi 0.80.6: max thinking level and input-based pricing tiers

Adds an opt-in max thinking level above xhigh (natively supported on GPT-5.6 and adaptive Claude models) across CLI, SDK, RPC, and model selection, plus request-wide input-token pricing tiers for accurate long-context cost accounting, configurable for custom models. Notable for builders tracking long-context costs or wanting more reasoning headroom.

PUBLISHED2026-07-09
OBSERVED2026-08-21
AGE1mo
SOURCES1
  • New opt-in max thinking level above xhigh; supported natively on GPT-5.6 and adaptive Claude models; available via CLI (--thinking max), SDK, RPC, and model selection. Custom themes can define thinkingMax, falling back to thinkingXhigh.
  • Request-wide input-token pricing tiers for accurate long-context cost accounting (e.g. GPT-5.4/5.5/5.6 long-context rates); configurable for custom models in models.json and modelOverrides.
  • Added ~ (home directory) expansion for the shellPath setting.
  • Fixed inherited post-compaction output-token budgeting ignoring stale pre-compaction assistant usage.
  • Fixed GPT-5.4/5.5 long-context cost accounting while keeping the intentional 272K default context limit.
  • Fixed GPT-5.6 metadata (272K short-context tier for direct OpenAI requests, 372K via Codex backend with long-context pricing) and removed the nonexistent bare gpt-5.6 alias.
  • Fixed Anthropic message conversion to preserve thinking blocks with empty thinking text but a valid signature.
featurecapabilitypricingfixmodel-support

COMMUNITY

No curated reactions recorded for this event. Facts and takes are kept in separate layers — community context is added by hand, never blended into the record above.