deepseek-v4.1POST
DeepSeek V4.1 Flash launches with DeepSWE 77.0, open weights and search API
The post says the new model scores 77.0 on DeepSWE, leads on Terminal-Bench, uses fewer tokens, ships open weights and opens a $0.75/1K-call search API to Claude Code and Codex.
良竹
@liangzhu_AIArchived 10 days agoOriginally posted 2026.10.17
良竹 @liangzhu_AIOriginally posted 2026.10.17
卧槽，那头蓝鲸又杀回来了！ DeepSeek-V4.1-Flash-1017 刚刚突袭上线，看完基准测试图直接倒吸一口凉气 🐋💥 不仅在 DeepSWE（77.0）和 Terminal-Bench 上把一众对手按在地上摩擦，干活更利索、废话更少、更省 Token 钱！ 顺手还开源了模型权重，甚至把自家搜索引擎 API（$0.75/千次）都开放给 Claude Code 和 Codex 用了。 诚意拉满，国产开源之光！👍
This post quotes another post: x.com/deepseek_ai/status/2111375640717033777
x.com/liangzhu_ai/status/2111383118594863480Archived 2026.10.18 16:48