DeepSeek V4 ships as public preview April 24 — dual-MoE release with V4-Pro at 1.6T total params and V4-Flash at 284B, both 1M native context
DeepSeek V4 launched in public preview on April 24, 2026, with two MoE models: V4-Pro at 1.6T total parameters and V4-Flash at 284B, both supporting a 1M-token native context window. The dual-tier release pattern mirrors the closed-source frontier-lab structure (premium + flash variants) for the first time in the open-source category.
The substantive piece is the open-source category maturing structurally. Pre-2026 open-source releases were monolithic — one model, one parameter count, take it or leave it. DeepSeek's V4 dual-MoE release pattern matches the closed-source frontier-lab structure (Claude Sonnet+Opus+Fable, GPT premium+mini, Gemini Pro+Flash) and lets enterprise procurement run the same per-workload cost-optimization analysis on open-source models that they already run on closed-source. The 1M-context window across both tiers is the floor that open-source releases will be expected to clear from here on.
The competitive read against Llama 4 Scout's 10M context is that the open-source long-context-window leaderboard now has clear separation: Scout at 10M, V4-Pro/Flash at 1M, the rest of the open frontier at 200K-512K. The deployment-shape segmentation is also clean — Scout for ultra-long-document workloads on a single H100, V4 for cost-optimized cluster deployments where context length and inference cost both matter.
LLM Stats — AI Updates Today (June 2026) – Latest AI Model Releases → · AI Unpacking — Open Source AI Models 2026: Llama, Mistral, DeepSeek & The Complete Guide →