### What happened?

Azure deployments of **gpt-5.5 and the gpt-5.6 family** reject ANY non-default `temperature` (e.g. 0.3) with a 400, but litellm forwards the value — and proxy-level `litellm_settings.drop_params: true` does **not** strip it, because the param-dropping layer only removes params the model map says are *unsupported*, and the map/GPT-5 config treat temperature as supported for these models.

Systematic sweep (2026-07-14) of every Azure chat model on our proxies: only gpt-5.5 and gpt-5.6-{luna,sol,terra} reject non-default temperature; gpt-5.4, o3, o4-mini and everything else accept 0.3. Working per-deployment workaround: `litellm_params.additional_drop_params: ["temperature"]`.

On current `litellm_oss_daily_2026_07_13`, `OpenAIGPT5Config._map_openai_params` forwards flexible temperature whenever the model `_supports_reasoning_effort_level(model, "none")` and no effort is set — which holds for gpt-5.5+ — so the request still 400s against Azure even with `drop_params` enabled (the drop branch is unreachable in that case).

### Repro

```python import litellm litellm.drop_params = True litellm.completion( model="azure/gpt-5.5", # Azure deployment of gpt-5.5 …