### What happened?

A single **non-streaming** `POST /v1/messages` (the dedicated Anthropic Messages endpoint) emits **two** OpenTelemetry spans named `litellm_request` for the one request. Both spans carry identical input/output/token-usage and identical duration, but **different** (regenerated) response ids (`chatcmpl-...`). The success path runs twice as well, so the cost/spend logging and any registered success callback fire twice per request.

A plain `POST /chat/completions` against the same proxy/config does **not** double — the issue is specific to the `/v1/messages` (`anthropic_messages`) path.

This is the **non-streaming** counterpart of the streaming pass-through duplication fixed in #29550 / #29598. That fix is guarded by `logging_obj.stream is True`, so it does not cover this non-streaming case.

### Version

`v1.82.3` (also present on `main` at time of filing).

### Minimal reproduction

`config.yaml`:

```yaml model_list: - model_name: claude litellm_params: model: anthropic/claude-3-5-sonnet-latest api_key: os.environ/ANTHROPIC_API_KEY

litellm_settings: success_callback: ["otel"] ```

Env: `OTEL_EXPORTER_OTLP_ENDPOINT=...` (any OTLP backend; a …