Chinese AI lab DeepSeek released two preview versions of its newest large language model, DeepSeek V4, on April 24, 2026, marking a significant update to last year’s V3.2 model and its accompanying R1 reasoning model.
The two new releases — DeepSeek V4 Flash and V4 Pro — are both mixture-of-experts models featuring context windows of 1 million tokens each, large enough to accommodate extensive codebases or documents within a single prompt. The mixture-of-experts architecture activates only a subset of parameters per task, which lowers inference costs.
V4 Pro is the larger of the two, with 1.6 trillion total parameters and 49 billion active parameters, making it the biggest open-weight model currently available. That figure surpasses Moonshot AI’s Kimi K 2.6 (1.1 trillion parameters), MiniMax’s M1 (456 billion), and more than doubles DeepSeek’s own V3.2 (671 billion). The smaller V4 Flash carries 284 billion total parameters with 13 billion active.
DeepSeek says both models outperform V3.2 through architectural improvements and have nearly “closed the gap” with leading models on reasoning benchmarks. The company claims its V4-Pro-Max variant beats open-source peers on reasoning tasks and surpasses OpenAI’s GPT-5.2 and Gemini 3.0 Pro on some evaluations. Coding performance for both V4 models is described as “comparable to GPT-5.4.” However, the models trail frontier models in knowledge tests — specifically GPT-5.4 and Google’s Gemini 3.1 Pro — a gap DeepSeek estimates reflects a “developmental trajectory that trails state-of-the-art frontier models by approximately 3 to 6 months.”
Both models support text only, unlike many closed-source competitors that handle audio, video, and image inputs and outputs.
Pricing is notably lower than comparable offerings. V4 Flash costs $0.14 per million input tokens and $0.28 per million output tokens, undercutting GPT-5.4 Nano, Gemini 3.1 Flash, GPT-5.4 Mini, and Claude Haiku 4.5. V4 Pro is priced at $0.145 per million input tokens and $3.48 per million output tokens, undercutting Gemini 3.1 Pro, GPT-5.5, Claude Opus 4.7, and GPT-5.4.
The launch arrives one day after the U.S. accused China of stealing American AI labs’ intellectual property at scale using thousands of proxy accounts. DeepSeek itself has faced separate accusations from Anthropic and OpenAI of “distilling” — essentially copying — their AI models.
Source: TechCrunch