AI Gateway production index

Vercel's AI Gateway analyzes production AI model usage across 200K+ teams, revealing how different models compete for different workload layers.

• •Anthropic leads in spending with 61% share due to high-value reasoning tasks, while Google dominates token volume (38%) with cost-efficient models
• •Agentic workloads carry 59% of all tokens (up 2x in 6 months), with tool-call requests being 2.6× more token-heavy than standard chat
• •Production teams at scale use 35+ distinct models as a multi-model routing architecture, making vendor lock-in less relevant and upgrades quicker
• •New model versions absorb market share within weeks of release, with Claude Sonnet 4.6 and Opus 4.7 rapidly displacing predecessors
• •Provider outages trigger fallbacks on 3.5% of requests (5.1% by tokens), with expensive long-context and reasoning calls more vulnerable to failures