DeepSeek V4.1 Flash 1017 对比 GPT-5.6 Sol 与 Opus 5：三项跑分与 20 倍价
称 DeepSWE 77.0、Terminal-Bench 4.0 42.3、Automation-Bench 60.5，三项均超 GPT-5.6 Sol、两项超 Opus 5，单价低至 1/20，另有 MIT 权重与 1M 上下文。
🚨 DeepSeek just released V4.1-Flash-1017 and it could become a serious headache for OpenAI and Anthropic. It's reportedly beating GPT-5.6 Sol on all 3 of these benchmarks and Opus 5 on 2. And it's up to 20x CHEAPER per token. → DeepSeek Flash: $0.30 / $1.20 per million tokens at peak. → GPT-5.6 Sol: $4 / $20. → Opus 5: $5 / $25. The benchmark scores: → DeepSWE: 77.0 → Terminal-Bench 4.0: 42.3 → Automation-Bench: 60.5 It also comes with open weights, an MIT license, and a 1M-token context window. Obviously, benchmarks aren't everything. Real-world reliability matters. But if Flash delivers on real coding tasks, why pay 10-20x more? Would you trust it to run your coding agents?
这条是引用着另一条写的：x.com/deepseek_ai/status/2111375640717033777