V4.1 Flash 1017 对比 Opus 5：DeepSWE 77.0 对 74.0，Terminal-Bench
列出 V4.1 Flash 1017 对 Opus 5 的跑分：DeepSWE 77.0 对 74.0、Automation-Bench 60.5 对 50.3，但 Terminal-Bench 4.0 与 HLE 仍落后；附 1M 上下文、图像输入、MIT 权重和 API 价格。
You don't always need an expensive AI model for coding. DeepSeek V4.1 Flash 1017 just posted some impressive benchmark numbers against Opus 5. • DeepSWE: 77.0 vs 74.0 • Automation-Bench: 60.5 vs 50.3 • 1M context window • Image input support • MIT license Opus 5 still leads on Terminal-Bench 4.0 and HLE reasoning, so it's not a clean sweep. But an open-weight model with low API costs putting up these numbers is worth watching. Try it free:http://chat.deepseek.com Open weights: http://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash API pricing: $0.15 input / $0.60 output per 1M tokens. Free access to the older V4.1 Flash: https://build.nvidia.com/ https://opencode.ai/ Open-source AI keeps getting more competitive.