| 规格 | 公布价值 |
|---|---|
| 创客 | 深度搜索 |
| 骨干网参数 | 552B |
| 有效参数 | 8B预填充/16B解码 |
| 语境 | 最多 100 万个代币 |
| 输入→输出 | 图片和文字 → 文字 |
| 建筑学 | 40层因果编码器-解码器;教育部 |
| 执照 | 麻省理工学院;型号卡和重量 |
制造商模型卡于 2026 年 10 月 30 日检查。主干网数量与总存储库张量参数不同。 · 来源
Unofficial, from StatusGator. Uptime is the share of the window not marked down (164 min down, 290 min degraded); it covers the whole service, not this version alone.
Estimate from the version's published specs, benchmarks and reviews, 0-5; scores are only comparable within one criteria year. Scored as launched (DeepSeek-V4.1-Flash, 10 Sep 2026). Reasoning: Artificial Analysis Intelligence Index 39 (v4.3.2, Max). Coding: Terminal-Bench 4.0 26.8% measured by AA (DeepSeek's own headline is Terminal-Bench 2.1 90.6, not comparable). Context 1M (pricing page). Price: DeepSeek API peak $0.30 in / $1.20 out (AA lists $0.295/$1.20; off-peak half) = about $0.53 blended. Speed: AA 181 t/s on its page today, 227 on 23 Sep per a report, 190-270 across variants; scored 4.5. Modalities: text and image in, text out. The pin's V4-Pro traffic routes to Flash. Reviews: AA model page, TNW, SiliconANGLE, buildfastwithai (blog); no independent hands-on found.[4]
It's also notably fast, however very verbose.
The open question is whether DeepSeek's benchmarks hold up in independent testing, and what V4.1-Pro costs when it arrives.
V4.1-Flash scored 90.6 on Terminal-Bench 2.1, narrowly ahead of Opus 5 at 89.1 and GPT-5.6 Sol at 88.8.
V4.1 Flash is one of the strongest value-focused models for coding agents, multimodal developer workflows and high-volume API use.
| DeepSeek launches V4.1-Flash and retires V4-Pro, its flagship model · 2026年9月10日 | 0.00 |
|---|---|
| DeepSeek releases V4.1-Flash, says it outperforms flagship V4-Pro · 2026年9月10日 | 0.70 |
| DeepSeek V4.1 Flash Review: Benchmarks, Price, Speed & Is It Worth It? (2026) · 2026年9月10日 | 0.70 |
这个标记会跟进新闻:它显示目前已知的情况,这里列出改了什么以及起因。
读者上传的照片,与图钉带来源的媒体分开显示。你可以为吃到或看到的内容评分,在餐厅图钉上还能直接在照片上标记菜品。
第一项始终是图钉的来源。总体可信度是各份资料对上方所用开始和结束时间支持程度的加权平均;资料每比最新的一份旧 180 天,权重减半。
DeepSeek[3][4] 的公告日期为“2026 年 9 月 10 日”,文中写道:“V4.1-Flash 现已在 DeepSeek API 上线”;其 API 文档更新日志中列有“DeepSeek-V4.1-Flash 发布 2026/09/10”,路透社也报道 DeepSeek“今天(9 月 10 日)推出了 DeepSeek-V4.1-Flash”。
Wayback capture dated 2026-09-11 of the original Artificial Analysis page; it reads "It's also notably fast, however very verbose", index 40, 250M tokens generated against a 130M median. Archive capture date, not the review's own publication date.
DeepSeek's[1][4] pricing page gives DeepSeek-V4.1-Flash a 1M context length and 384K maximum output, and per-million-token prices of $0.15 input (cache miss) and $0.60 output off-peak, $0.30 and $1.20 at peak, with peak hours 01:00-04:00 and 06:00-10:00 UTC on weekdays.
DeepSeek's[1][3] developer-docs copy of the announcement, listed as "DeepSeek-V4.1-Flash Release 2026/09/10": the model name deepseek-flash, the retirement of V4-Flash and V4-Flash-Vision-Exp, and "Starting at 04:00 UTC on Sept 14, 2026, all deepseek-v4-pro requests will route to V4.1-Flash at V4.1-Flash rates".
The official model card: 552B backbone parameters, a Causal Encoder-Decoder that activates 8B parameters per token in prefill and 16B in decode, a KV cache of 890 bytes per token, 45T pre-training tokens, the MIT license, and the benchmark table against Opus 5.0, GPT-5.6 Sol, Kimi K3 and GLM-5.3.[1]
有遗漏或错误吗?用你自己的话说明:能佐证此图钉的链接、不同的开始或结束日期及理由,或缺失、有误的信息。AI 会对照此图钉的来源进行核实,搜索更好的来源,并添加任何支持你说法的页面。图钉自身的来源仍然最重要。AI 也会查看图片:显示的是别的东西或显示效果差的图片会被移到后面或替换。
原标题: DeepSeek-V4.1-Flash Released
deepseek-flash 之名在 DeepSeek API 中上线,并接入 DeepSeek 的网页版和移动应用[1][6]deepseek-v4-pro 请求都由 V4.1-Flash 处理,并按 V4.1-Flash 的价格计费,直到 V4.1-Pro 发布为止[1][4]Epoch AI model database (DeepSeek V4.1 Flash); Epoch's confidence in the figures: Likely. Blank tiles are not published.