open-sourceDeepSeek·Released 2024-12
DeepSeek V3
DeepSeek's flagship open-weights MoE model. Chosen when price and open weights matter more than vendor reputation.
Pricing verified May 7, 2026 · source
Key specifications
| Vendor | DeepSeek |
| Tier | open-source |
| Context window | 128K tokens |
| Pricing | Open weights; very low API pricing — see api-docs.deepseek.com |
| Released | 2024-12 |
| Parameters | 671B (MoE, ~37B active) |
| Categories | Text GenerationCode GenerationAdvanced ReasoningOpen Source |
Note: api-docs.deepseek.com/quick_start/pricing now lists only deepseek-v4-flash, deepseek-v4-pro, and deepseek-v4-flash-vision-exp. The deepseek-chat and deepseek-reasoner (V3) model IDs were retired 2026-07-24 15:59 UTC. V3 weights remain accessible via third-party hosts such as OpenRouter but are no longer served by DeepSeek's own API. Entry left unchanged per PR #59 precedent.
Key features
- •Open weights
- •Mixture-of-Experts — large total / small active
- •Very competitive on coding and math benchmarks
- •Aggressive API pricing
Compare with similar models
Model lineups change often. AI Flash Report tracks new model releases and maintains a deprecation calendar for models being retired.
← Back to all AI models