DeepSeek V4—almost on the frontier, a fraction of the price

DeepSeek has released two preview models in their V4 series: DeepSeek-V4-Pro (1.6T total parameters, 49B active) and DeepSeek-V4-Flash (284B total, 13B active), both with 1M token context windows and MIT licensing. The standout feature is pricing — V4-Flash at $0.14/$0.28 per million tokens is the cheapest small model available, undercutting GPT-5.4 Nano, while V4-Pro at $1.74/$3.48 is the cheapest large frontier model. Efficiency gains from the architecture explain the low cost: V4-Pro uses only 27% of the FLOPs and 10% of the KV cache compared to V3.2 at 1M-token context. DeepSeek's own benchmarks show V4-Pro competitive with frontier models but trailing GPT-5.4 and Gemini-3.1-Pro by roughly 3–6 months. Both models are available on Hugging Face, with quantized versions from Unsloth expected soon.