Interesting:

• Qwen3 beats Gemma-3
• Gemini-2.5-pro beats Claude-4-Sonnet
• this too shows the instruction regression between o3 and gpt-5
• gpt-oss-120b scores really high in this benchmark ^[1]

1. though for my personal use-cases it was worse than qwen3 and gemini-2.5-flash! ↩