DeepSeek V3
DeepSeek V3 was an open-weight large language model from DeepSeek that drew significant attention for its strong performance at a low training and inference cost. DeepSeek has since moved to its V4 model family (V4.1-Flash and V4-Pro), with V3's legacy API model names retired in July 2026.
Facts checked on against the sources listed at the end of this page.
Newer version available. DeepSeek's legacy model API names, including those tied to the V3 generation, were retired on July 24, 2026, with traffic routed to the newer V4 family. DeepSeek V4 is the current model family. See DeepSeek V4.
- Pricing
- Open sourcePaid from Superseded by DeepSeek V4 API pricing
- Open source
- YesMIT (weights and code)
- GitHub stars
- 104,507
- Latest release
- v1.0.01 year ago
A good fit for
- Researching the history of DeepSeek's model releases and their impact on open-weight AI
- Self-hosting the V3 open weights for projects that specifically require that version
Not a good fit for
- New API integrations, since DeepSeek's legacy V3-era API model names have been retired in favor of the V4 family
- Anyone wanting DeepSeek's current best-performing model, where V4 is the active generation
DeepSeek V3 pricing
DeepSeek V3's weights are free to download and self-host under the MIT license. Its API model names have been retired in favor of the DeepSeek V4 family; see DeepSeek V4 for current API pricing.
Prices in USD, checked on 27 September 2026. Current prices on deepseek.ai.
Main features
- Open weights
- Released under the MIT license, allowing self-hosting and modification.
- Cost-efficient training and inference
- Notable for strong performance relative to its training and serving cost.
- API and third-party hosting
- Available through DeepSeek's own API and various inference providers.
About DeepSeek V3
DeepSeek V3 was an open-weight large language model that gained widespread attention for achieving strong benchmark performance while being trained and served at a notably lower cost than many proprietary competitors.
As an open-weight model, it could be downloaded and self-hosted in addition to being accessed through DeepSeek's own API and various third-party inference providers.
DeepSeek has since released its V4 model family, including V4.1-Flash and V4-Pro, and retired the legacy API model names associated with the V3 generation as of July 24, 2026, routing existing traffic to the newer models.
- Made by
- DeepSeek
- Free plan
- Yes, open-weight model free to download and self-host; API access is pay-as-you-go
- Available on
- API, Self-hosted (open weights), Various inference providers
Usage and activity
Recent releases
- v1.0.01 year ago
DeepSeek V3 alternatives
Questions about DeepSeek V3
Is DeepSeek V3 still available?
The open-weight model files remain downloadable, but DeepSeek retired the legacy API model names tied to the V3 generation on July 24, 2026 in favor of the V4 family.
What replaced DeepSeek V3?
The DeepSeek V4 family, including V4.1-Flash and V4-Pro.
Sources and links
Work on DeepSeek V3? Claim this listing. Something wrong or out of date? Suggest a correction. Stats come from the public GitHub and npm APIs and update every day.