### The Feature

LiteLLM does not support **HTTP/2 for outbound requests** to upstream LLM providers.

When LiteLLM acts as a client calling provider APIs (Bedrock, Vertex, Anthropic, OpenAI, etc.), the httpx clients are created **without** `http2=True`:

- `litellm/llms/custom_httpx/http_handler.py:392` — `httpx.AsyncClient(...)` created without `http2=True` - `litellm/llms/custom_httpx/http_handler.py:943` — `httpx.Client(...)` created without `http2=True`

Since httpx defaults to `http2=False`, **all outbound calls use HTTP/1.1**. A repo-wide search for `http2=True` returns zero matches.

Additionally, the default async transport is **aiohttp** (`_should_use_aiohttp_transport()` returns `True` by default), and aiohttp does not support HTTP/2 at all — so even if the httpx flag were set, the default code path would still be HTTP/1.1.

> Note: The `h2` package present in the dependency tree is a transitive dependency of **hypercorn** (used for the *inbound* server side via `--run_hypercorn`, which already documents "supports HTTP/2"). It is not wired into the outbound client.

**Request:** Add opt-in HTTP/2 support for outbound provider requests, e.g. a `litellm.enable_http2` flag …