1.7.3

Keeps up with the newest models, and stops a stalled response from hanging the caller.

The newest models do not all take reasoning effort NONE. OpenAI gpt-6.1-sol and gpt-6-astra, and Anthropic Claude Opus 5.5 and Claude Fable, cannot turn reasoning off at all, and asking for it answered HTTP 400. On OpenAI this hit every plain chat call, as the default effort AUTO asks for none there. These models now fall back to LOW, the lowest effort they accept. Claude Sonnet 5.5 turns thinking off with a thinking.type of its own, between_tools, which is now sent for it (#26, #28).

A response body which stalls after its headers no longer blocks the caller forever. The request timeout of the JDK HTTP client covers the headers alone, so a provider which answered 200 and then stopped sending held the call indefinitely. The body now has to arrive within the request timeout as well, and a stalled one is retried like any other timeout. An error response whose body stalls still throws the exception of its status. This was seen on OpenRouter, where a model now and then keeps generating for minutes while the connection waits (#27).

An interrupt while reading a response body is now reported as an interrupt, and the interrupt flag of the thread is kept (#25).

The default models move up: gpt-6.1-sol for OpenAI and Azure OpenAI, claude-sonnet-5-5 for Anthropic, gemini-3.8-flash for Google AI, and grok-4.7 for xAI.

SmallRye Config 4.0 works as MicroProfile Config implementation, and the README now names it.