DeepSeek-V3-0324 · 官方评测披露

DeepSeek-V3-0324 当前收录 4 项测试的官方记录。已完成列明材料的核查。

模型发布 · DeepSeek V3

官方材料与核查范围

All listed official model-card, official News/API, metadata and directly linked technical-report materials for this release were read through the requested scope. Tables without an explicit unit retain an empty unit; figures without exact labels were not estimated. 核对时间 。

当前资料不足以还原首发。 No reliable historical snapshot proving the complete launch-day material set was obtained. Official publication dates and current materials do not substitute for a launch snapshot.

  • DeepSeek-V3-0324 official Hugging Face model card · 主文 · HF revision e9b33add76883f293d6bf61f6bd89b497e80e335; current card snapshot
    公布 未确认 · 采集 · 已核查 Official model-card Markdown read in full; benchmark rows, notes, configuration statements and linked reports checked.
  • DeepSeek-V3-0324 official Hugging Face model metadata · 主文 · author=deepseek-ai; repo createdAt=2025-03-24T09:28:22.000Z (repository upload time only); lastModified=2025-03-27T04:01:53.000Z; sha=e9b33add76883f293d6bf61f6bd89b497e80e335
    公布 未确认 · 采集 · 已核查 Official model API metadata checked for author, repository identity, revision and timestamps; createdAt was never used as release date.
  • DeepSeek-V3 模型更新,各项能力全面进阶 · 主文 · Official DeepSeek News index entry dated 2025-03-25; index downloaded and retained locally
    公布 · 采集 · 已核查 The official dated entry or current service table was read for this release scope; aliases were not treated as weight checkpoints.

配置

  • DeepSeek-V3-0324 · The release overview gives model-to-model improvement values; per-row benchmark settings are not stated.
    未知项 No test version or metric is inferred where absent from the quoted line.

直接前代与后续发布

直接前代 DeepSeek-V3。 官方关系依据

DeepSeek-V3-0324 demonstrates notable improvements over its predecessor, DeepSeek-V3, in several key aspects.
0 项披露延续 · 4 项新增 · 34 项本次未见 · 0 项待核查
DeepSeek-V3 与 DeepSeek-V3-0324 的披露变化
测试已核查状态DeepSeek-V3DeepSeek-V3-0324
MMLU-Pro · V3-0324 overview does not state metric or set version新增—本次记录
GPQA · V3-0324 overview does not state variant or metric新增—本次记录
AIME · Year/metric not stated in V3-0324 overview新增—本次记录
LiveCodeBench · Version/metric not stated in V3-0324 overview新增—本次记录
MMLU-Pro · DeepSeek card version unspecified; Exact Match本次未见前代记录—
GPQA Diamond · Diamond subset; task snapshot unspecified本次未见前代记录—
LiveCodeBench · Version unspecified; Pass@1-CoT本次未见前代记录—
Codeforces · Percentile; contest snapshot unspecified本次未见前代记录—
SWE-bench Verified · Agent mode; task snapshot unspecified本次未见前代记录—
AIME · 2024 edition; Pass@1本次未见前代记录—
Aider-Polyglot · Version unspecified; Accuracy本次未见前代记录—
MMLU · DeepSeek-V3 Technical Report Table 6; MMLU (EM); unit not printed本次未见前代记录—
MMLU-Redux · DeepSeek-V3 Technical Report Table 6; MMLU-Redux (EM); unit not printed本次未见前代记录—
DROP · DeepSeek-V3 Technical Report Table 6; DROP (3-shot F1); unit not printed本次未见前代记录—
IF-Eval · DeepSeek-V3 Technical Report Table 6; IF-Eval (PromptStrict); unit not printed本次未见前代记录—
SimpleQA · DeepSeek-V3 Technical Report Table 6; SimpleQA (Correct); unit not printed本次未见前代记录—
FRAMES · DeepSeek-V3 Technical Report Table 6; FRAMES (Acc.); unit not printed本次未见前代记录—
LongBench v2 · DeepSeek-V3 Technical Report Table 6; LongBench v2 (Acc.); unit not printed本次未见前代记录—
HumanEval-Mul · DeepSeek-V3 Technical Report Table 6; HumanEval-Mul (Pass@1); unit not printed本次未见前代记录—
LiveCodeBench · DeepSeek-V3 Technical Report Table 6; LiveCodeBench (Pass@1); unit not printed本次未见前代记录—
Aider-Edit · DeepSeek-V3 Technical Report Table 6; Aider-Edit (Acc.); unit not printed本次未见前代记录—
MATH-500 · DeepSeek-V3 Technical Report Table 6; MATH-500 (EM); unit not printed本次未见前代记录—
CNMO · DeepSeek-V3 Technical Report Table 6; CNMO 2024 (Pass@1); unit not printed本次未见前代记录—
CLUEWSC · DeepSeek-V3 Technical Report Table 6; CLUEWSC (EM); unit not printed本次未见前代记录—
C-Eval · DeepSeek-V3 Technical Report Table 6; C-Eval (EM); unit not printed本次未见前代记录—
C-SimpleQA · DeepSeek-V3 Technical Report Table 6; C-SimpleQA (Correct); unit not printed本次未见前代记录—
Arena-Hard · DeepSeek-V3 Technical Report Table 7; Arena-Hard; unit not printed本次未见前代记录—
AlpacaEval 2.0 · DeepSeek-V3 Technical Report Table 7; AlpacaEval 2.0 (length-controlled win rate); unit not printed本次未见前代记录—
RewardBench · DeepSeek-V3 Technical Report Table 8; Chat; unit not printed本次未见前代记录—
RewardBench · DeepSeek-V3 Technical Report Table 8; Chat-Hard; unit not printed本次未见前代记录—
RewardBench · DeepSeek-V3 Technical Report Table 8; Safety; unit not printed本次未见前代记录—
RewardBench · DeepSeek-V3 Technical Report Table 8; Reasoning; unit not printed本次未见前代记录—
RewardBench · DeepSeek-V3 Technical Report Table 8; Average; unit not printed本次未见前代记录—
RewardBench · DeepSeek-V3 Technical Report Table 8; Chat; maj@6; unit not printed本次未见前代记录—
RewardBench · DeepSeek-V3 Technical Report Table 8; Chat-Hard; maj@6; unit not printed本次未见前代记录—
RewardBench · DeepSeek-V3 Technical Report Table 8; Safety; maj@6; unit not printed本次未见前代记录—
RewardBench · DeepSeek-V3 Technical Report Table 8; Reasoning; maj@6; unit not printed本次未见前代记录—
RewardBench · DeepSeek-V3 Technical Report Table 8; Average; maj@6; unit not printed本次未见前代记录—

测试记录 · 4