Go SDK 0.10.2
One slow model provider response no longer fails the whole run. OpenAI, Anthropic and Google model calls default to a ten-minute request timeout bounded by the run's deadline, and retry up to twice with jittered backoff on provider timeouts and 408/429/5xx/529, honouring Retry-After. This matches the Python and TypeScript SDKs, including the AGNT5_LM_MAX_RETRIES, AGNT5_LM_INITIAL_DELAY_MS and AGNT5_LM_MAX_DELAY_MS overrides. A request that still fails returns a ModelRequestError that says whether the provider timed out (AGNT5-1251). The worker also reports its service version correctly again.
No breaking API changes.