FrontierMay 12, 2026via Vercel Blog
AI Gateway production index
Why it matters
Vercel's anonymized production data from 200K+ teams reveals that AI provider dominance is use-case-fragmented, not monolithic. Anthropic dominates high-stakes spend, Google leads volume on cheap inference, and teams at scale route across 35+ models, making traditional model benchmarks obsolete for real-world decision-making.
Key signals
- April 2026 spend: Anthropic 61%, Google 21%, OpenAI 12%
- April 2026 token volume: Google 38%, Anthropic 26%, OpenAI 13%, xAI 10%
- Agentic workloads: 22.2% of requests, 58.9% of tokens (2.6x more token-heavy than non-agentic)
- Agent tool-call share doubled from 11.4% (Oct 2025) to 22.2% (Apr 2026)
- High-volume teams (10M+ requests) use average 35 distinct models in regular rotation
- Claude Sonnet 4.6 absorbed most of Sonnet family share within 1 month of launch
- Claude Opus 4.7 migration mirrors Sonnet 4.6 adoption curve
- Anthropic 71% token share in back-office, drops to 7% in consumer
- Google Gemini Flash: 28% of consumer tokens at 15% of cost
- B2B applications spend roughly 2x per token vs B2C
- 3.5% of requests complete via fallback routing; 5.1% by tokens; 4.9% by cost
- OSS models (Kimi, MiniMax, GLM) significant in consumer and building tiers
- Data source: 7 months Vercel AI Gateway production traffic, 200K+ unique teams, through April 2026
The hook
Anthropic 61% of spend, Google 38% of volume. The AI market isn't one race—it's five happening at once, and your model choice depends on how expensive being wrong is.
Ask which AI model is best, and the answer changes before the ink dries. That's what happens in an industry where new models are released weekly.
Every benchmark measures a different race, and every race crowns its own winner, but Vercel has a unique view of the industry through production workload…