AI.info
DeepSeek Flash — Model Tracker
DeepSeek Flash in the Model Tracker: every value read on the maker's own pages, with its day.
- Maker
- DeepSeek — huggingface.co, read 2026-10-16
- Released
- 2026-09-10 — www.deepseek.com, read 2026-10-16
- Status
- Generally available — www.deepseek.com, read 2026-10-16
- Retired
- Not confirmed
- Takes
- Text, Image — huggingface.co, read 2026-10-16
- Returns
- Text — huggingface.co, read 2026-10-16
- Open weights
- Yes — huggingface.co, read 2026-10-16
- Licence
- MIT License — huggingface.co, read 2026-10-16
- Context window
- 1,000,000 tokens — huggingface.co, read 2026-10-16
- Longest answer
- 384,000 tokens — api-docs.deepseek.com, read 2026-10-16
- Knowledge cutoff
- Not confirmed: The reviewed official model card and documentation do not state a knowledge cutoff.
- API price, input
- Not confirmed: The official pricing page lists different input prices by cache status and peak/off-peak period, rather than one standard input price.
- API price, output
- Not confirmed: The official pricing page lists different output prices for peak and off-peak periods, rather than one standard output price.
- Where
- API, Downloadable weights — www.deepseek.com, read 2026-10-16
- Parameters
- Not confirmed: two readings disagree: 763000000000 and 552000000000
- Model card
- https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash — huggingface.co, read 2026-10-16
Sources
- DeepSeek-V4.1-Flash: Pushing the Limits of KV Cache Compression (read 2026-10-16)
- Introducing DeepSeek-V4.1-Flash: smarter, faster, more efficient. (read 2026-10-16)
- Models & Pricing (read 2026-10-16)