DeepSeek V4 - Almost on the Frontier, a Fraction of the Price

Metadata

• Author: Simon Willison’s Weblog
• Full Title: DeepSeek V4 - Almost on the Frontier, a Fraction of the Price
• URL: https://simonwillison.net/2026/Apr/24/deepseek-v4/#atom-everything

Highlights

• Chinese AI lab DeepSeek’s last model release was V3.2 (and V3.2 Speciale) last December. They just dropped the first of their hotly anticipated V4 series in the shape of two preview models, DeepSeek-V4-Pro and DeepSeek-V4-Flash. (View Highlight)
• Both models are 1 million token context Mixture of Experts. Pro is 1.6T total parameters, 49B active. Flash is 284B total, 13B active. They’re using the standard MIT license. (View Highlight)
• I think this makes DeepSeek-V4-Pro the new largest open weights model. It’s larger than Kimi K2.6 (1.1T) and GLM-5.1 (754B) and more than twice the size of DeepSeek V3.2 (685B). (View Highlight)
• Pro is 865GB on Hugging Face, Flash is 160GB. I’m hoping that a lightly quantized Flash will run on my 128GB M5 MacBook Pro. It’s possible the Pro model may run on it if I can stream just the necessary active experts from disk. (View Highlight)
• So the pelicans are pretty good, but what’s really notable here is the cost. DeepSeek V4 is a very, very inexpensive model. (View Highlight)
• Here’s DeepSeek’s pricing page. They’re charging $0.14/milliontokensinputand$0.28/million tokens output for Flash, and $1.74/millioninputand$3.48/million output for Pro. (View Highlight)
• DeepSeek-V4-Flash is the cheapest of the small models, beating even OpenAI’s GPT-5.4 Nano. DeepSeek-V4-Pro is the cheapest of the larger frontier models. (View Highlight)
• In the scenario of 1M-token context, even DeepSeek-V4-Pro, which has a larger number of activated parameters, attains only 27% of the single-token FLOPs (measured in equivalent FP8 FLOPs) and 10% of the KV cache size relative to DeepSeek-V3.2. Furthermore, DeepSeek-V4-Flash, with its smaller number of activated parameters, pushes efficiency even further: in the 1M-token context setting, it achieves only 10% of the single-token FLOPs and 7% of the KV cache size compared with DeepSeek-V3.2. (View Highlight)
• I’m keeping an eye on huggingface.co/unsloth/models as I expect the Unsloth team will have a set of quantized versions out pretty soon. It’s going to be very interesting to see how well that Flash model runs on my own machine. (View Highlight)