The November 2026 SuperCLUE benchmark of 19 models ranks Google's Gemini-4-Argon first at 73.62, with Moonshot's Kimi-K3.1-Pro the leading open model at 61.46.
Key points
- GPT-6.1-Sol is second at 71.76; DeepSeek-V4.1-Flash 58.95; Zhipu GLM-5.3 50.66.
- Gemini leads long context (95.63) and hallucination control (83.19); GPT-6.1-Sol tops math reasoning (76.32).
- Kimi's edge is hallucination control (67.78), the best among domestic models.
Why it matters
The gap between frontier closed models and open weights remains wide, but open models are closing the reliability gap faster than the raw score gap suggests.
Read the full post with the infographic: https://www.threerouter.com/blog/317
Sources: threerouter AI Daily.
Top comments (0)