Validate small high-scoring models

[EwoutH]

It would be great if all high-scoring small models on MMLU-Pro could be validated to provide reliable and complete scores. These small models are valuable as they're fast and cheap to run while showcasing important trends in model and distillation efficiency.

Small, high-scoring models

QwQ Family

• QwQ-32B-Preview (32B)
• QwQ-32B (32B)

Microsoft/Phi Family

• Phi-4 (14B)
• Phi-4-mini (5.6B)
• Phi3-medium-4k (14B)

Qwen Family

• Qwen2.5-32B (32B)
• Qwen2.5-14B (14B)

Google Family

• Gemma-3-27B-it (27B)
• Gemma-3-12B-it (12B)
• Gemma-2-27B-it (27B)

Mistral Family

• Mistral-Small-instruct (24B)
• Mistral-Small-base (24B)

Other Models

• SkyThought-T1 (32B)
• Reka 3 (21B)
• RRD2.5-9B (9B)
• EXAONE-3.5-32B-Instruct (32B)
• Internlm3-8B-Instruct (8B)