DeepSeek

DeepSeek changed expectations around model efficiency and cost with R1 in January 2025, and it remains one of the most important Chinese frontier labs. V4 Pro (1M context) delivers low-cost reasoning and coding at roughly a tenth of US frontier prices; V4.1 Flash adds image input at $0.30 per million input tokens.

• DeepSeek V4.1 Flash (Sep 10, 2026): budget model with image input, 1M context.
• DeepSeek V4 Pro (Aug 13, 2026): $1.32/$3.96 per million tokens at peak, half off-peak.
• Open weights continue to be downloaded and fine-tuned worldwide.

DeepSeek proves frontier-class capability can be cheap and open. Every price cut it makes ripples through the whole market.

Editorial score, 0–100. How it works →