llmix's 2.2.0 patch fixes a silent reasoning-effort bug

llmix silently dropped the reasoningEffort parameter for provider-prefixed model names, per the project's GitHub release notes for version 2.2.0. The failure showed up on a model addressed as openai/gpt-5.6-luna, whose prefix broke the router's anchored capability matching. Pricing lookup for that model came back empty too. Neither failure threw an error.

Anyone who sets an effort level in code assumes it reaches the model. Here it didn't, and nothing in the response flagged the drop. OpenRouter and LiteLLM route models the same prefixed way, so the same gap can show up there too. Checking that the effort level actually reaches the provider takes a few minutes.

Version 2.2.0 is where the fix lands, tagged on GitHub. Anyone still on an earlier build hasn't gotten it, and any provider-prefixed model string is worth a second look.

If effort levels have ever quietly not worked, don't assume the model ignored the setting. Check the actual request your gateway sent before it left.