Model dossier模型档案
DeepSeek V4 Flash 0423
DeepSeek V4 Flash 0423 is a high-tier model from DeepSeek (Intelligence Score 83.5/100). Strongest in agent (97.1) and math (97); best fit: tool-use and multi-step agents. It trails in preference (53). Budget-priced: $0.036 in / $0.071 out per 1M tokens. Context window 1M tokens — long-document friendly. Open weights — self-hostable. Reachable via CN-direct endpoints.
DeepSeek V4 Flash 0423 是 DeepSeek 的高水平梯队模型(智能评分 83.5/100)。 最强项是智能体(97.1分),其次是数学(97分);适合工具调用与多步 Agent。 短板是偏好(53分)。 定价属低价档:每百万 token 输入 $0.036 / 输出 $0.071。 上下文 1M token,长文档友好。 开放权重,可自托管。 支持国内直连。
open weights开放权重 ⌂ CN direct国内直连
83.5
Intelligence Score · benchmark coverage 智能评分 · benchmark 覆盖 88%
◈
Benchmark scoresBenchmark 成绩
9 entries 条| Benchmark基准 | Score成绩 | Index指数 | Dated日期 | Measured by测评方 | |
|---|---|---|---|---|---|
| Humanity's Last ExamReasoning | 32.1 | 56 | — | 3rd-party第三方 | source ↗×2⚠±17.3 |
| GPQA DiamondReasoning | 87.4 | 92 | — | 3rd-party第三方 | source ↗×5⚠±3.9 |
| MMLU-ProKnowledge | 86.2 | 93 | 2026-08 | 3rd-party第三方 | source ↗×5⚠±18.1 |
| SWE-bench VerifiedCoding | 79 | 82 | 2026-08 | 3rd-party第三方 | source ↗×5⚠±5.3 |
| AIME 2025Math | 97 | 97 | — | 3rd-party第三方 | source ↗×3⚠±27 |
| LiveCodeBenchCoding | 91.6⚡max | 98 | — | 3rd-party第三方 | source ↗×5⚠±26.1 |
| ↳ ⚡high | 88.4 | — | — | 3rd-party第三方 | source ↗ |
| Terminal-BenchCoding | 82.7 | 90 | 2026-07-31 | 3rd-party第三方 | source ↗×3⚠±25.8 |
| τ²-benchAgent | 95.6⚡high | 97 | 2026-06-10 | 3rd-party第三方 | source ↗×3 |
| LMArena (Chatbot Arena)Preference | 1435 | 53 | 2026-08-12 | official官方榜 | source ↗×4⚠±41 |
≡
Specs & pricing规格与价格
| Provider厂商 | DeepSeek |
|---|---|
| Released发布 | 2026-04-24 |
| Context window上下文 | 1M tokens |
| Max output最大输出 | 384K tokens |
| Modality模态 | text->text |
| Reasoning tiers推理档位 | ⚡none ⚡high ⚡max default默认 high · thinking off / high effort / max effort |
| API input price输入价格 | $0.036 / 1M tokens |
| API output price输出价格 | $0.071 / 1M tokens |
⛓
Tools using this model使用它的工具
⚔
Head-to-head正面对比
How we score →评分方法 → · Catalog & pricing via the official OpenRouter API.模型目录与价格来自 OpenRouter 官方 API。