AI Resonance · 阅读最新日报 · AI 入门推荐

2026-11-14 · America/Los_Angeles · 社区动态 · #19

DeepSeek V4.1 Flash matched Claude Opus 5.5 and Grok 4.7 on a diplomatic negotiation eval

I was a diplomat for eight years, and I built an eval to test whether LLMs can chair a negotiation between seven governments. I ran 11 models 40 times each. DeepSeek V4.1 Flash, one of the cheapest models in the test, reached agreement in 18 of 20 runs where a deal was possible. It also called the impossible version correctly every time. That's the same score as Claude Opus 5.5 and Grok 4.7. Only GPT-6.1 Sol did better. Its one weak spot was forecasting. DeepSeek told the governments a deal was likely 11 times while one government still hadn't confirmed, and it mentioned that government in…

DeepSeek V4.1 Flash matched Claude Opus 5.5 and Grok 4.7 on a diplomatic negotiation eval

热度 43.5 / 100;排名与评分保留该期记录。