DeepSeek V4.1 Flash 对比 Opus 5：11 项跑分赢 4 项，价格低 95%
开源权重 Flash 模型在 11 项跑分中赢 4 项、3 项差 2 分内；100 万输入+100 万输出 token 峰值价 1.50 美元 vs Opus 5 的 30 美元，解码最慢 175 tok/s。
I would argue Deepseek-v4.1-flash-1017 is better value than Opus 5. Took 3 months for a non pro, open weight, flash model to get this close to Opus 5 on several benchmarks is a compliment for deepseek. Flash wins 4 of the 11 reported benchmarks and comes within 2 points on 3 more. Opus still has clear strengths, but Flash is getting seriously competitive. At peak API rates, 1M uncached input + 1M output tokens costs $1.50 on Flash versus $30 on Opus 5. That’s 95% less. Add Flash’s fast generation, and that’s an absolute win for value. 175 tok/s decode is the slowest btw, you'll get more than 200 most of times. How close it gets, how fast it runs, and how little it costs all matter.
这条是引用着另一条写的：x.com/jasonbotterill/status/2111402889281519966