Analyst: V4.1 is now the best Chinese model, far more compute-efficient on ARC benchmarks
teortaxesTex · x · 2026-10-08
Analyst teortaxesTex reconstructs compute usage from Opus and argues V4.1 is likely more compute-efficient on ARC-2 and vastly more efficient on ARC-3, making it the best Chinese model on ARC-2 given Dots' pricier attention and similar active params. He is also lobbying ARC to display token use by default.
Related event: ARC-2 Data Suggests Zhipu GLM V4.1 May Be China's Strongest Model(2 posts)→
More from Models
- Musk praises Claude Haiku 4.5 as Anthropic's cheapest, fastest small model, ~75% cheaper — elonmusk · 2026-10-08
- Perplexity releases pplx-embed-v2-late: open late-interaction embeddings for text, image and pages — antoine_chaffin · 2026-10-08
- Vik Paruchuri calls out model copying: inspiration doesn't let you rename it and change the license — VikParuchuri · 2026-10-08
- Haiku 5.5 costs 20x less than Sonnet 5.5, dev uses it for parallel sub-agents — RLanceMartin · 2026-10-08
- 100 classic texts tested: Pangram flags Victor Hugo poem as 100% AI after one line removed — GolinoHudson · 2026-10-08
- Haiku 5.5 at max reasoning burns more tokens than Opus 5.5 max, user finds — haider1 · 2026-10-08