docs: explain model capability overrides

[neubig]

What changed

• Document provider-neutral reasoning_effort values and LiteLLM translation.
• Explain metadata-first capability detection for new and proxied models.
• Show how to use model_canonical_name for gateway aliases.
• Document the api_mode and capability_overrides escape hatches.
• Clarify that extended_thinking_budget is only for confirmed manual-thinking models, not adaptive-thinking models.

Why

OpenHands/software-agent-sdk#4200 generalizes model capability resolution so new models can be enabled through metadata instead of SDK model-name allowlists. Users of custom gateways need a concise explanation of the automatic path and the explicit fallback controls.

Validation

• git diff --check
• Docs tests: 34 passed. Two unrelated pricing tests error because their existing https://raw.githubusercontent.com/OpenHands/OpenHands/main/openhands/utils/llm.py fixture returns 404.

Dependency

• fix(llm): generalize model capability resolution software-agent-sdk#4200

[mintlify]

Preview deployment for your docs. Learn more about Mintlify Previews.

Project	Status	Preview	Updated (UTC)
all-hands-ai	🟢 Ready	View Preview	Jul 22, 2026, 8:17 PM

💡 Tip: Enable Workflows to automatically generate PRs for you.