DeepSeek-V3-0324: the quiet update that puts vendor lock-in on notice
In one sentence DeepSeek releases a DeepSeek-V3 update (685B param MoE, 37B active) under MIT license. Performance close to Claude 3.7 Sonnet on coding, training cost estimated 20x lower.
Two months after R1 (January), DeepSeek quietly drops — no press release — an update to its flagship DeepSeek-V3. The file appears on Hugging Face on March 24: 685 billion total parameters, only 37 billion active per answer ("Mixture of Experts" architecture).
The key point: it's free. MIT license, downloadable, commercial use allowed. And on coding and math benchmarks it gets close to Claude 3.7 Sonnet, which costs $3/$15 per million tokens. DeepSeek offers it via API at $0.27/$1.10.
Effect: cloud providers (Together, Fireworks, OpenRouter) add it within 48 hours. Companies using Claude or GPT-4o for coding seriously start evaluating the Chinese open-weight alternative.
Companies
DeepSeek
Tools
DeepSeek-V3-0324
Tags
Sources