正在读取模型档案与评测数据…
正在读取模型档案与评测数据…
DeepSeek·4 个数据来源·更新于 2026-09-26
只在同一来源、同一口径下比较。切换来源可查看不同测试结果。
当前配置：各基准设置见下表
综合分数和同榜图表来自 LLM Stats 榜单（采集于 2026-09-25）；下方逐项成绩来自模型页。榜单出处
LLM Stats Score · 同一来源与榜单口径，分数越高越好
| 模型与配置 | LLM Stats Score | 每任务成本 |
|---|---|---|
| GLM-5.3 | 52.1 | 未公布 |
| Atria Dawn Preview | 51.6 | 未公布 |
| Qwen3.8 Max | 51.5 | 未公布 |
| DeepSeek-V4.1-Flash · 当前模型 | 51.4 | 未公布 |
| DeepSeek-V4-Pro-0813 | 51.1 | 未公布 |
| Hy4 preview | 50.4 | 未公布 |
不同来源的分数不能直接相加或横向比较；评测任务成本不是 API Token 单价。
20 项已收录成绩 · 保留原始单位和测试条件
Pass@1; official scaffold. Maximum reasoning effort (reasoning_effort=100); temperature=1.0; top_p=0.95.
成绩出处Pass@1; official scaffold. Maximum reasoning effort (reasoning_effort=100); temperature=1.0; top_p=0.95.
成绩出处Pass@1; with tools; Claude Code harness; 512K-token context. Maximum reasoning effort (reasoning_effort=100); temperature=1.0; top_p=0.95.
成绩出处Pass@1; with tools; Claude Code harness; 512K-token context. Maximum reasoning effort (reasoning_effort=100); temperature=1.0; top_p=0.95.
成绩出处Rating. Maximum reasoning effort (reasoning_effort=100); temperature=1.0; top_p=0.95.
成绩出处Pass@1. Maximum reasoning effort (reasoning_effort=100); temperature=1.0; top_p=0.95.
成绩出处Resolved; mini-SWE harness; 1M-token context. Maximum reasoning effort (reasoning_effort=100); temperature=1.0; top_p=0.95.
成绩出处Pass@1. Maximum reasoning effort (reasoning_effort=100); temperature=1.0; top_p=0.95.
成绩出处Pass@1; GPQA Diamond. Maximum reasoning effort (reasoning_effort=100); temperature=1.0; top_p=0.95.
成绩出处Pass@1; full multimodal evaluation, no tools. Maximum reasoning effort (reasoning_effort=100); temperature=1.0; top_p=0.95.
成绩出处Pass@1; text-only subset, no tools. Maximum reasoning effort (reasoning_effort=100); temperature=1.0; top_p=0.95.
成绩出处Pass@1; with tools; full multimodal evaluation. Maximum reasoning effort (reasoning_effort=100); temperature=1.0; top_p=0.95.
成绩出处Pass@1. Maximum reasoning effort (reasoning_effort=100); temperature=1.0; top_p=0.95.
成绩出处Score; DeepSeek Harness Minimal; 1M-token context. Maximum reasoning effort (reasoning_effort=100); temperature=1.0; top_p=0.95.
成绩出处Almost@1; DeepSeek Harness Minimal; 1M-token context. Maximum reasoning effort (reasoning_effort=100); temperature=1.0; top_p=0.95.
成绩出处Pass@1; Claude Code harness. Maximum reasoning effort (reasoning_effort=100); temperature=1.0; top_p=0.95.
成绩出处Pass@1; DeepSeek Harness Minimal; 1M-token context. Maximum reasoning effort (reasoning_effort=100); temperature=1.0; top_p=0.95.
成绩出处Pass@1; DeepSeek Harness Minimal; 1M-token context. Maximum reasoning effort (reasoning_effort=100); temperature=1.0; top_p=0.95.
成绩出处Pass@1; DeepSeek Harness Minimal; 1M-token context. Maximum reasoning effort (reasoning_effort=100); temperature=1.0; top_p=0.95.
成绩出处Pass@5; main split with tools; Claude Code harness; 512K-token context. Maximum reasoning effort (reasoning_effort=100); temperature=1.0; top_p=0.95.
成绩出处只展示有可靠来源的资料；优先采用厂商官网与官方模型仓库。
美元 / 百万 Token。不同服务商或测试配置的报价分别列出。
| 来源 / 服务商 | 输入 | 输出 | 缓存读取 |
|---|---|---|---|
| Fireworks | $0.22 | $0.66 | $0.007 |
| DeepInfra | $0.3 | $1.2 | $0.006 |
| DeepSeek | $0.3 | $1.2 | $0.006 |
| Novita | $0.3 | $1.2 | $0.006 |
同名模型通过开发机构和明确版本关联。预览版、日期版本和不同规模模型分别建档。
公开评测数据；保留来源原始口径
公开模型页与公开榜单；自报成绩单独标识
Arena 官方公开数据集 · CC BY 4.0
公开模型目录与 API 报价
| 来源 / 配置 | 采集时间 | 原始数据 |
|---|---|---|
| Artificial AnalysisDeepSeek V4.1 Flash (max)SHA-256 b9324ff6388c2fd3… | 2026-09-25T19:40:27Z | 查看来源 |
| Artificial AnalysisDeepSeek V4.1 Flash (Non-reasoning)SHA-256 b9324ff6388c2fd3… | 2026-09-25T19:40:27Z | 查看来源 |
| Artificial AnalysisDeepSeek V4.1 Flash (Reasoning, Max Effort)SHA-256 5beb2f0dcdc468c1… | 2026-09-26T00:27:47Z | 查看来源 |
| Artificial AnalysisDeepSeek V4.1 Flash (Non-Reasoning)SHA-256 5beb2f0dcdc468c1… | 2026-09-26T00:27:47Z | 查看来源 |
| LLM StatsDeepSeek-V4.1-Flash | 2026-09-25T19:40:20Z | 查看来源 |
| LLM StatsDeepSeek-V4.1-FlashSHA-256 04db0831bfd9b185… | 2026-09-26T00:28:56Z | 查看来源 |
| ArenaDeepseek V4.1 Flash (Max) | 2026-09-25T19:40:08Z | 查看来源 |
| OpenRouterDeepSeek V4.1 FlashSHA-256 4c26b1452fbd2acd… | 2026-09-26T00:28:52Z | 查看来源 |
