Commit Graph
3 Commits
Author SHA1 Message Date
46e6c01a01 fix(claude): reconcile max_tokens vs thinking budget and lift per-model ceiling (#2381)
On the translated OpenAI->Claude path, adjustMaxTokens capped max_tokens
before applyThinking set thinking.budget_tokens, so max-effort budget
(128000) could exceed a 64k-clamped max_tokens -> Anthropic 400.
prepareClaudeRequest now reconciles after the budget is known: prefer
raising max_tokens, only shrink budget when it meets/exceeds the ceiling.

Also lift the global 64000 cap: the ceiling is now the model's real
maxOutput, so high-output models (fable/mythos, opus-4.8/sonnet-4.6) get
their full budget. adjustMaxTokens gains an optional ceiling arg (default
unchanged, callers untouched); openai-to-claude passes the model maxOutput.

Native Claude Code passthrough is unaffected.

Co-Authored-By: Claude <noreply@anthropic.com>
Co-authored-by: Cursor <cursoragent@cursor.com>
2026-07-05 17:38:16 +07:00
decolua d03f9fb823 Enhance configuration and model capabilities 2026-06-16 23:32:28 +07:00
decoluaandCursor d3f61aac2f refactor(open-sse): translator DRY + schema enums, bug fixes, dead code cleanup
- Bug B1-B7: media UI m.kind||m.type, serviceKinds, gemini mediaPriority, schema kind, models/info lookup by kind
- Dead code D1-D6: safeParseJSON, drop PROVIDER_ENDPOINTS, orphan fetcher, GITHUB_CONFIG derive, getProviderConfig internal, legacy kiro file
- Translator concerns: toOpenAIUsage, toOpenAIFinish (gemini/kiro/ollama + fix kiro tool finish), thinking effort maps
- Reorg helpers/ → concerns/ (logic) + formats/ (per-format) + schema/ (pure enums: roles/blocks/finishReasons/defaults)
- Wire ~280 hardcoded role/block/finish/default literals to schema enums across 20+ files
- collapseTextParts + extractTextContent dedup
- Normalize translator fn names to openaiToXRequest / xToOpenAIResponse
- Golden tests lock behavior; 0 regression (byte-for-byte providers/alias, 26=26 known fails)

Co-authored-by: Cursor <cursoragent@cursor.com>
2026-06-14 18:49:38 +07:00