### What happened?

VibeThinker models from WeiboAI are not currently supported in litellm, despite being popular reasoning-optimized small models.

**Models to add:** - `WeiboAI/VibeThinker-1.5B` (1.5B params, based on Qwen2.5-Math-1.5B) - 1,003 downloads, 535 likes on HuggingFace - Released: November 2025 - Paper: https://arxiv.org/abs/2511.06221 - `WeiboAI/VibeThinker-3B` (3B params, based on Qwen2.5-Coder-3B) - 6,589 downloads, 372 likes on HuggingFace - Released: June 2026 - Paper: https://arxiv.org/abs/2606.16140

**Key features:** - Reasoning-optimized (math, code, GPQA) - MIT licensed - Compatible with HuggingFace Inference API and vLLM - OpenAI-compatible API via text-generation-inference

**Use case:** Small, efficient reasoning models for cost-effective inference with strong performance on mathematical and logical reasoning tasks.

### What should have happened?

Users should be able to call VibeThinker models via litellm:

```python import litellm

response = litellm.completion( model="huggingface/WeiboAI/VibeThinker-3B", messages=[{"role": "user", "content": "Solve: 2x + 5 = 13"}] ) ```

### Implementation notes

Since VibeThinker models are Qwen…