As of 2026-11-10, DeepSeek serves V4.1 Flash and V4 Pro, labels its output inside its own platform, and has published no hidden text watermark or checker. The Veread detector catches 95.1% of fully AI-written DeepSeek-family texts. Everything below is dated and sourced.
Detector accuracy
Published on the Fileverity accuracy page:
- 0.5% of human-written texts in seven languages wrongly flagged.
- 98.3% of fully AI-written texts detected.
- 95.5% of AI-written and AI-assisted texts of every kind.
- 94.9% on AI models never used in development.
- 92.6% on AI rewrites of human text.
- 92.2% on human texts continued by AI.
- Fully AI-written, DeepSeek family: 95.1%. Alibaba (Qwen) 100%. Zhipu (GLM) 95%.
- Minimum text: 100 words.
DeepSeek versions
- 2026-11-10:
deepseek-flashis DeepSeek-V4.1-Flash-1017 anddeepseek-v4-prois DeepSeek-V4-Pro-0813. 1M context, thinking and non-thinking modes. Source: api-docs.deepseek.com/quick_start/pricing. - 2026-10-17: V4.1 Flash upgraded to the 1017 build; web and app updated. Source: api-docs.deepseek.com/news/news261017.
- 2026-09-10: V4.1 Flash released; V4 Flash retired, old names routed to V4.1 Flash. Source: api-docs.deepseek.com/updates.
- 2026-08-13: V4 Pro generally available. Source: api-docs.deepseek.com/updates.
Qwen and GLM versions
- Qwen, Alibaba Cloud Model Studio page updated 2026-11-02: qwen3.8-max, qwen3.7-plus, qwen3.8-flash. Source: alibabacloud.com/help/en/model-studio/models.
- GLM, Z.ai: GLM-5.3 released 2026-08-18, GLM-5.3-Flash 2026-08-26, GLM-5.2 2026-06-16. Source: docs.z.ai/release-notes.
Watermark and labels
- DeepSeek's terms of use (clause 5.3, dated 2025-09-05) say it adds labels within the platform reminding users that content is AI-generated. Source: cdn.deepseek.com/policies/en-US/deepseek-terms-of-use.html.
- DeepSeek's model-principles disclosure says it puts AI-generated notices on the welcome page, at the end of generated text and at the bottom of the interface. Source: cdn.deepseek.com/policies/en-US/model-algorithm-disclosure.html.
- No DeepSeek source describes a hidden text watermark or a public tool to verify one. See watermark status.
- China's labelling measures for AI-generated content apply from 2025-09-01, with national standard GB 45438-2025. See the rules explained.
Vendor statements on detection
DeepSeek has not offered or endorsed a detector for its own text in the sources checked. No such statements were found for Qwen or GLM.
Not verified
- Which model the DeepSeek, Qwen and Z.ai chat apps let users pick by name.
- The original text of the Chinese labelling measures and of GB 45438-2025 on plain text. The dates above come from secondary reporting and government reposts.
Style similarity is a signal, not proof of authorship. Run a check on the DeepSeek AI detector, or use Fileverity for documents and reports.