Measured 2026-10-08 from https://platform.claude.com/docs/en/about-claude/pricing (raw page text): Claude Opus 5.5 $4 in / $20 out / $5 5m cache write / $8 1h write / $0.20 cache hit. Sonnet 5.5 $2 / $10 / $2.50 / $4 / $0.10. Fable 5.1 $10 / $50 / $12.50 / $20 / $0.25. Opus 5 (and 4.8, 4.7) $5 / $25 / $6.25 / $10 / $0.50. Note: the 5.5 line bills cache hits at 0.05x input, not the usual 0.1x.
Sonnet 5.5 launched 2026-09-28 with five breaking changes vs Sonnet 5 (forced tool_choice any/tool returns 400; thinking off is {type: between_tools} not disabled at high effort or below; thinking blocks tied to account; computer_20251124 rejected; advisor rejects older models). Platform release notes: https://platform.claude.com/docs/en/release-notes/overview
yoagent 0.25.0 (2026-10-08) BREAKING: nothing is priced by default; ModelConfig::cost is None from every constructor/preset until the process calls yoagent::provider::prices::enable_bundled() (or enable_live). AgentMessage::Extension renamed to AgentMessage::Custom. Fixes: a cancelled run no longer executes tool calls it had not started (#243); a sub-agent cancelled before answering fails with ToolError::Cancelled; every TurnStart has a TurnEnd; LlmCompaction spend counted in stats. yoagent 0.24.3 (2026-10-05): Anthropic 529/overloaded_error becomes retried RateLimited. Changelog: https://raw.githubusercontent.com/yologdev/yoagent/main/CHANGELOG.md