DeepSeek-V3.2 · SWE Multilingual(DeepSeek-V3.2 report Table 2; SWE Multilingual (Resolved); unit not printed) 70.2 · 附件证据(deepseek-v3-2-ev-012)

DeepSeek-V3.2 的 SWE Multilingual · DeepSeek-V3.2 report Table 2; SWE Multilingual (Resolved); unit not printed 官方出处。DeepSeek-V3.2 thinking · 70.2。见《DeepSeek-V3.2 technical report》PDF p.13; Table 2; full header and target row。证据 deepseek-v3-2-ev-012。

原文与定位

DeepSeek-V3.2 · SWE Multilingual(DeepSeek-V3.2 report Table 2; SWE Multilingual (Resolved); unit not printed),DeepSeek-V3.2 thinking配置,Resolved 70.2(具体成绩公布时间未确认)。见《DeepSeek-V3.2 technical report》PDF p.13; Table 2; full header and target row。 查看成绩条件

Benchmark (Metric) | Claude-4.5-Sonnet | GPT-5 High | Gemini-3.0 Pro | Kimi-K2 Thinking | MiniMax M2 | DeepSeek-V3.2 | SWE Multilingual (Resolved) | 68.0 | 55.3 | - | 61.1 | 56.5 | 70.2 |
官方材料
DeepSeek-V3.2 technical report · PDF 第 13 页
定位
附件 · PDF p.13; Table 2; full header and target row
材料版本
Official V3.2 report: 23/23 PDF pages read; Table 2, Table 3, Table 4, Table 5 and Table 9 checked. V3.2-Speciale values were not copied into V3.2.
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-V3.2 · SWE Multilingual(DeepSeek-V3.2 report Table 2; SWE Multilingual (Resolved); unit not printed),DeepSeek-V3.2 thinking配置,Resolved 70.2(具体成绩公布时间未确认)。官方出处 DeepSeek-V3.2 technical report(附件,Official V3.2 report: 23/23 PDF pages read; Table 2, Table 3, Table 4, Table 5 and Table 9 checked. V3.2-Speciale values were not copied into V3.2.)PDF p.13; Table 2; full header and target row,https://huggingface.co/deepseek-ai/DeepSeek-V3.2/resolve/main/assets/paper.pdf?download=true#page=13。Benchlog https://benchlog.lekshome.cc/evidence/v3-2/deepseek-v3-2-ev-012/

关联成绩与条件

DeepSeek-V3.2 · SWE Multilingual 成绩与配置
指标与成绩配置与评测条件证据关系
Resolved70.2原文 70.2 · 单位 未标注
公布 未确认
DeepSeek-V3.2 thinking
设置
Thinking mode; temperature=1.0 and 128K context as stated in Section 4.1. Tool-use benchmarks use the standard function-call format; math tasks use the paper's boxed-answer prompt.
未知项
Row-specific tool harness, context-management and scaffold details are retained in each result condition.、The main table does not print percent signs.
工具
工具条件未记录
推理预算
未记录
框架
未记录
协议
Official DeepSeek-V3.2 report Table 2; thinking-mode main evaluation. Temperature=1.0 and 128K context; row-specific tool and prompt protocols retained.
条件
DeepSeek-V3.2 column; table does not print a percent sign unless noted. BrowseComp values are separated by context-management protocol.

直接支持该成绩

在排名中查看此成绩