DeepSeek enters the fight for token volume, Anthropic continues to dominate spend
17%. That's how much token share DeepSeek V4 captured in a single month—without taking a dime from Anthropic's spend dominance.
Why it matters
DeepSeek V4's aggressive pricing ($0.14/$0.28 vs. Anthropic's $2-5 range) has fractured the AI market into a two-tier system: cheap, capable models for volume work and frontier models for high-stakes tasks. This signals a fundamental shift in how teams architect production AI—and where AI lab revenue actually comes from.
The key facts
11 to knowDeepSeek V4 jumped from <1% to 17% token share in one month (April to May 2026)
DeepSeek V4 Flash priced at $0.14 input / $0.28 output per million tokens (20-50x cheaper than Anthropic, 8-12x cheaper than Qwen/Kimi)
Despite 17% token share, DeepSeek captured only ~1% of spend
Anthropic grew spend share from 61% to 65% and holds 70-80% across high-stakes use cases
OpenAI token share flat at 13%, but customers paying more per token (spend share 12% to 13%)
Total AI Gateway tokens grew +20% MoM; total spend grew +43% MoM
Average token cost increased ~20% despite DeepSeek's low-cost entry
Tool-call requests carry >50% of tokens despite being <25% of requests
Gemini 3.5 Flash adoption stalled (only 7% of Flash family) vs. rapid 3.1 Pro migration
B2B applications cost 60% more per token than B2C
Apps handling 1M+ requests use 11+ models in production; lowest-volume tier dominated by single-model setups
Go to the source
Vercel Blogvercel.com
Publisher excerpt: Every month, routes tens of trillions of tokens between production applications and AI labs, giving us visibility into what AI usage actually looks like, separate from leaderboards and benchmarks. We publish the data monthly in the AI Gateway production index. AI Gateway Last month, headlines about…