SDK disables httpx's default timeout

[mdrxy]

When timeout_ms is not explicitly configured (the default), the SDK passes timeout=None to httpx.Client.build_request(), which overrides httpx's built-in Timeout(5.0) default and disables all timeouts for the request. This causes indefinite hangs when TCP connections stall during response body reads.

Root cause

basesdk.py:212:

timeout = timeout_ms / 1000 if timeout_ms is not None else None

This None is then passed to:

client.build_request(
    ...,
    timeout=timeout,  # None → disables httpx timeout entirely
)

In httpx, an explicit timeout=None means "no timeout" — it does not mean "use the client default." The correct sentinel for "no override" is httpx.USE_CLIENT_DEFAULT.

Impact

• sdk_configuration.timeout_ms defaults to None
• Every request made without an explicit timeout_ms has no read timeout
• If a TCP connection stalls (e.g., CloudFlare → provider network issue), the request blocks forever
• We've reproduced this at ~5% rate against nvidia/nemotron-3-super-120b-a12b via DeepInfra — requests hang indefinitely during receive_response_body while OpenRouter logs show the generation completed normally

Reproduction

import openrouter

sdk = openrouter.OpenRouter(api_key="...")
print(sdk.sdk_configuration.timeout_ms)  # None

# This request has NO timeout — will hang forever on TCP stall
response = sdk.chat.send(messages=[...], model="nvidia/nemotron-3-super-120b-a12b")