## Summary

LiteLLM's Opus 4.7/4.8 model detection does not tolerate the Vertex version suffix (`@default` etc.). This causes the model to be misidentified when used with the recommended model ID, which blocks configuration of adaptive thinking parameters and effort levels.

The model-name fallbacks in `litellm/llms/anthropic/common_utils.py` (specifically `_is_claude_4_7_model()`, the `opus_4.7` membership check in `_is_adaptive_thinking_model()`, and the missing explicit check for 4.8) recognize the bare name but fail when Vertex appends a version suffix.

## Evidence

```python _is_adaptive_thinking_model("vertex_ai/claude-opus-4-8@default") -> False # WRONG _is_adaptive_thinking_model("vertex_ai/claude-opus-4-8") -> True # OK

_model_supports_effort_param("claude-opus-4-8@default") -> False # WRONG _model_supports_effort_param("claude-opus-4-8") -> True # OK ```

Vertex routinely appends `@<version>` (e.g. `@default`, `@20251101`) to the model identifier, and that suffix survives provider-prefix stripping, so the detection helpers see `claude-opus-4-8@default` and fail to recognize it as an adaptive-thinking model.

The 4.6 name-based…