DeepSeek-V4-Pro (Preview) · R&D Coding Benchmark (internal)(30-task internal engineering evaluation; Pass Rate (%)) 67% · 附件证据(deepseek-v4-pro-preview-ev-034)

DeepSeek-V4-Pro (Preview) 的 R&D Coding Benchmark (internal) · 30-task internal engineering evaluation; Pass Rate (%) 官方出处。DeepSeek-V4-Pro (Preview) Max · 67%。见《DeepSeek-V4 technical report》PDF p.44; Table 8; full header and target row。证据 deepseek-v4-pro-preview-ev-034。

原文与定位

DeepSeek-V4-Pro (Preview) · R&D Coding Benchmark (internal)(30-task internal engineering evaluation; Pass Rate (%)),DeepSeek-V4-Pro (Preview) Max配置,Pass Rate 67%(具体成绩公布时间未确认)。见《DeepSeek-V4 technical report》PDF p.44; Table 8; full header and target row。 查看成绩条件

Model | Haiku 4.5 | Sonnet 4.5 | DeepSeek-V4-Pro-Max | Opus 4.5 | Opus 4.6 Thinking | Opus 4.6 Thinking Pass Rate (%) | 13 | 47 | 67 | 70 | 73 | 80 Pass Rate (%) | 13 | 47 | 67 | 70 | 73 | 80
官方材料
DeepSeek-V4 technical report · PDF 第 44 页
定位
附件 · PDF p.44; Table 8; full header and target row
材料版本
Official V4 report: 58/58 PDF pages read; Table 1, Table 6, Table 7, Table 8 and internal evaluation tables were checked. Preview values were not propagated to 0731/0813.
版本公布
未确认
披露时间
未确认 · 首发/后补未确认
采集时间
历史依据
暂无可靠历史版本依据
引用文本

DeepSeek-V4-Pro (Preview) · R&D Coding Benchmark (internal)(30-task internal engineering evaluation; Pass Rate (%)),DeepSeek-V4-Pro (Preview) Max配置,Pass Rate 67%(具体成绩公布时间未确认)。官方出处 DeepSeek-V4 technical report(附件,Official V4 report: 58/58 PDF pages read; Table 1, Table 6, Table 7, Table 8 and internal evaluation tables were checked. Preview values were not propagated to 0731/0813.)PDF p.44; Table 8; full header and target row,https://arxiv.org/pdf/2606.19348#page=44。Benchlog https://benchlog.lekshome.cc/evidence/v4-pro-preview/deepseek-v4-pro-preview-ev-034/

关联成绩与条件

DeepSeek-V4-Pro (Preview) · R&D Coding Benchmark (internal) 成绩与配置
指标与成绩配置与评测条件证据关系
Pass Rate67%原文 67 · 单位 %
公布 未确认
DeepSeek-V4-Pro (Preview) Max
设置
Official mode label: Max; the row-specific inference budget and sampling parameters are not stated in this model-card table.
未知项
Do not assume local recommended parameters are the evaluation settings.
工具
工具条件未记录
推理预算
Max
框架
未记录
协议
Official DeepSeek-V4 report Table 8; 30 internal R&D tasks with human-annotated rubrics.
条件
DeepSeek-V4-Pro-Max row; internal task set excluded from regular capability presentation.

直接支持该成绩

在排名中查看此成绩