[zannen7]

Summary

• Preserve ChatGPT Web's native thinking_effort=max tier.
• Map minimal/low/medium/standard -> standard, high/extended -> extended, and xhigh/max -> max.
• Keep the fix scoped to ChatGPT Web; the generic GPT-5.6 Responses max path is already handled upstream.
• Add focused regression coverage for both pure mapping and executor wire behavior.

Validation

• Provider change with automated regression test
• Reconciled with current release/v3.8.50 at 7938715
• Final diff reviewed; no dist/** or generated changes
• New focused test TypeScript syntax-transpiles with 0 diagnostics
• Focused provider gates / lint / SonarQube — CI pending

Tests Added Or Updated

• tests/unit/chatgpt-web-max-thinking-effort.test.ts: locks all eight mappings and verifies max/xhigh emit thinking_effort=max in both the settings PATCH and conversation body.

Reviewer Notes

• Builds on feat(chatgpt-web): support thinking_effort (Standard/Extended) for thinking-capable models #1821. high remains extended; only the highest alias xhigh follows the new highest web tier to max.
• Related to fix(providers): GPT-5.6 selected max effort rewritten to xhigh #8997 / fix(translator): thread model through normalizeResponsesReasoningEffort in promotion path (#8997) #9361; this PR does not duplicate the generic Responses fix.