Preview models will count

Other hosting providers (DeepInfra, for example) must be able to host the models themselves (the weights have to be available to download but vLLM or other engines don't have to support running it)

If any model of the family (e.g. DeepSeek v5 mini or DeepSeek v5 Flash) is released, then it counts.

The EST timezone will be used to decide if it releases before 2028 or not

Background

DeepSeek is an AI Lab based in China. It gained attention for training highly efficient open-weight models at a fraction of the cost compared to western labs like OpenAI or Anthropic. Following the success of DeepSeek-V3 and DeepSeek-R1, the company launched its DeepSeek-V4 models in 2026, bringing significant agentic and coding performance upgrades. As of August 2026, there is no official roadmap or timeline announced for DeepSeek V5.