v0.18.0

Highlights

The known-models capability table is refreshed to the current Anthropic / OpenAI / Gemini lineups (June 2026). Every model now callable through the three providers is enumerated with verified tool-use, vision, and per-model input-token budgets, researched against each provider's official docs.

Why

The knownModels table in internal/cli/capabilities.go gates which --llm-* selections are accepted and how tier validation reasons about a model's capabilities. As providers shipped new models and revised context windows, the table drifted — leaving newer models unrecognized and at least one stale context window on the books.

What changed

• Anthropic — adds claude-opus-4-8 plus the still-callable legacy/deprecated Claude 4.x rows.
• OpenAI — adds gpt-5.5-pro, gpt-5.4-pro, gpt-5.2, and gpt-5-nano; corrects gpt-5.4 to its now-published 1.05M context window (input budget 260k → 900k).
• Gemini — adds gemini-3-pro-preview, gemini-3-flash-preview, and the GA 2.5 family.

Changes were made test-first: capabilities_test.go gained lineup coverage asserting tool-use / vision flags and input budgets before the table was expanded.

What's Changed

• Refresh known-models table with current Anthropic/OpenAI/Gemini lineups by @britt in #110

Full Changelog: v0.17.1...v0.18.0