deepseek-v4.1IMAGE

DeepSeek V4.1-Flash-1017 released with open MIT weights, tops Opus 5 on DeepSWE

Official release: 763B params, 1M context, MIT open weights; DeepSWE 77.0 vs Opus 5's 74.0 at about $0.2 per task versus $10+, though it still trails on HLE and Terminal-Bench.

PANews丨APP全面升级

@PANewsCNArchived 10 days agoOriginally posted 2026.10.17

DeepSeek V4.1-Flash-1017 released with open MIT weights, tops Opus 5 on DeepSWE (image 1 of 2)

DeepSeek V4.1-Flash-1017 released with open MIT weights, tops Opus 5 on DeepSWE (image 2 of 2)

PANews丨APP全面升级 @PANewsCNOriginally posted 2026.10.17

Open original on X

DeepSeek开源V4.1-Flash更新版，多项编程和智能体测试超过Claude Opus 5 DeepSeek今天发布了V4.1-Flash-1017，这是V4.1-Flash的一次小版本更新，已在官网、App和API上线，模型权重也同步在Hugging Face以MIT许可证开源，参数规模7630亿，支持100万token上下文。官方称，新版本测试成绩更好，完成任务更快，输出也更短。在编程测试DeepSWE上，它拿到77.0分，高于Claude Opus 5的74.0和GPT-5.6 Sol的73.0；在Automation-Bench、Agents’ Last Exam等几项智能体测试中也排在第一。更夸张的是成本：DeepSWE上它每个任务平均只花约0.2美元，而Claude Opus 5要10美元以上。不过在HLE、Terminal-Bench等测试里，它仍落后于Claude Opus 5。

This post quotes another post: x.com/deepseek_ai/status/2111375640717033777

x.com/panewscn/status/2111381961361244291Archived 2026.10.18 16:45