DeepSeek announced DeepSeek-V4 Preview, describing V4-Pro and V4-Flash as live and open-sourced with 1M-token context as the default on official services, the company said on 24 April 2026.
Open-weight model labs are competing on long context and price as enterprises shop for reasoning APIs outside the US frontier set. Buyers still need workload benchmarks, licence terms and residency options.
DeepSeek publishes models and APIs from its official site. The V4 Preview post named API models deepseek-v4-pro and deepseek-v4-flash; this pack does not assert funding figures.
AI platform buyers should insist on reproducible evals on their prompts — before treating any long-context preview as a completed model-strategy decision.