23 days ago
- DeepSeek's V4 model family, released April 24, 2026, significantly increased the company's token share on OpenRouter, doubling from around 10% in January to nearly 20% by June.
- The surge in token share was driven by agentic workloads, which use about 15x more tokens per request than human AI usage, and V4 became the preferred model for such tasks, with V4-Flash comprising 70% of DeepSeek's agentic token flow by end of May.
- V4's cost-effectiveness, priced at $0.09/$0.18 per million tokens (input/output) compared to GPT-5.5's $5/$30, made it attractive for organizations, contributing to its growth without a proportional spike in spending share.
- The competitive landscape shifted in 2026, with Chinese models (led by DeepSeek, Xiaomi, Minimax, Tencent) surpassing American ones in token share, reversing 2025's dominance where U.S. models held about 75%.
- Usage growth spanned all user types, including hobbyists (nearly one-third of tokens routed to DeepSeek) and AI-native/large organizations, indicating broad adoption beyond cost-sensitive segments.
- The analysis is based on OpenRouter data from over 450 trillion tokens (Jan 1 - Jun 14, 2026), categorized into Agentic, Mixed, and Human workloads using a 7-signal composite score.