DeepSeek

DeepSeek is a Chinese AI company that achieved global prominence through its open-source large language models. As of April 2026, the flagship is the DeepSeek-V4 series, comprising DeepSeek-V4-Pro (1.6 trillion total parameters with 49 billion activated, Mixture-of-Experts) and DeepSeek-V4-Flash (284 billion total parameters with 13 billion activated), both supporting 1 million-token context windows. V4-Pro requires only 27% of the inference FLOPs and 10% of the KV cache of DeepSeek-V3.2 at the 1M-token setting, and DeepSeek prices V4-Pro at roughly $3.48 per million output tokens versus $25–$30 for comparable Western frontier models. Earlier models DeepSeek-V3, V3.2, and the reasoning-focused DeepSeek-R1 remain widely deployed. The full V4 weights are released under permissive open-source licenses, allowing developers to download, run locally, and fine-tune the model — a pattern that benefits Web3 builders deploying AI agents on decentralised compute and edge infrastructure. Example: DeepSeek-V4-Pro's 1M-token context and low cost make it practical for Web3 builders to feed entire smart-contract codebases plus governance forum history into a single inference call when auditing a protocol, a workflow that would be prohibitively expensive with closed frontier models. Why it matters for AI and data in Web3: DeepSeek's cost-efficient, open-source frontier models reduce barriers to deploying AI agents on decentralised compute networks (io.net, Akash, Render). Lower per-token economics make on-chain AI services and autonomous agent fleets viable that would be uneconomical on closed APIs.

Explore the full Web3 Glossary — 2,067+ expert-curated definitions. Need guidance? Talk to our consultants.