OpenAI recommends the Responses API (/v1/responses) over Chat Completions for new projects. This adds support for it across all OpenAI-related plugins while preserving existing Chat Completions behavior.

The Responses API uses input (array of role/content objects) instead of messages, and returns output array with structured content. Token usage fields differ: input_tokens/output_tokens vs prompt_tokens/completion_tokens.

Changes

OpenAIModels.cs

• Added OpenAIResponsesRequest, OpenAIResponsesResponse, and supporting types
• Extended TryGetOpenAIRequest to detect Responses API by checking for input array with objects containing role/type
• Added TryGetCompletionLikeRequest shared method for plugins that only handle text generation
• Extended OpenAIResponseUsage to deserialize both token naming conventions via property aliases

Plugin Updates

• LanguageModelFailurePlugin: Injects fault prompts into Responses API input array
• LanguageModelRateLimitingPlugin: Recognizes Responses API for rate limit tracking
• OpenAITelemetryPlugin: Adds telemetry tags/metrics for Responses API operations
• OpenAIMockResponsePlugin: Mocks Responses API by converting to/from chat completion format

Example Request Detection

// Responses API request structure
{
  "model": "gpt-4o",
  "input": [
    { "role": "user", "content": "Hello" }
  ]
}

// Detected via:
if (rawRequest.TryGetProperty("input", out var inputProperty) &&
    inputProperty.ValueKind == JsonValueKind.Array &&
    inputProperty.GetArrayLength() > 0)
{
    var firstItem = inputProperty.EnumerateArray().First();
    if (firstItem.TryGetProperty("role", out _) || firstItem.TryGetProperty("type", out _))
        // Responses API request
}

• Fixes Add support for OpenAI Responses API #1428

✨ Let Copilot coding agent set things up for you — coding agent works faster and does higher quality work when set up for your repo.