v0.4.0: OpenAI Responses API Support

🚀 New Features

OpenAI Responses API Support for Codex Models

This release adds full support for OpenAI's /v1/responses endpoint, enabling seamless integration with Codex models like gpt-5.1-codex.

Key Features:

• ✅ Automatic Detection: Codex models are automatically detected by name pattern
• ✅ Smart Routing: Codex models → /v1/responses, Standard models → /v1/chat/completions
• ✅ Transparent Client Experience: All requests still use /v1/messages endpoint
• ✅ Proper Transformations: Request/response formats converted automatically

📝 Changes

Request Handling

• Codex models use max_output_tokens instead of max_tokens
• Messages transformed to Responses API input format
• Full parameter mapping for Responses API

Response Parsing

• Parse Responses API output[] array structure
• Extract text from output_text content blocks
• Map input_tokens/output_tokens to Anthropic format

🔧 Implementation Details

New Types:

• OpenAIResponsesRequest - Request format
• OpenAIResponsesResponse - Response format
• ResponsesOutput - Output item structure
• ResponsesContentBlock - Content block structure

New Functions:

• is_codex_model() - Detect Codex models
• transform_to_responses_request() - Convert requests
• transform_responses_response() - Parse responses

✅ Testing

Successfully tested with:

• gpt-5.1-codex model
• Standard chat completion models (backward compatible)

📦 Files Changed

• src/providers/openai.rs: +244 lines, -33 lines

---

🤖 Generated with Claude Code