Vercel AI Gateway Data: Cheap Models Win Volume, But Expensive Models Drive Most Spend
Analysis of Vercel's AI Gateway reveals that while open-source models like DeepSeek dominate usage volume, closed-source frontier models like Claude account for the vast majority of API costs.
Practical Summary
This social media post shares data from Vercel's AI Gateway, showing a clear disconnect between model usage volume and associated cost. It provides a practical, data-backed perspective for optimizing AI tool spend by analyzing where money is actually flowing.
Why It Matters
For teams managing AI tool budgets, this highlights a critical optimization lever: the most popular models by volume are not necessarily the cost drivers. Understanding this pattern can guide routing decisions, model selection, and budget allocation to prioritize cost-effective solutions for specific tasks while reserving expensive models for high-value work.
Understanding the Volume vs. Cost Disconnect
The post cites data from Vercel's AI Gateway, which provides granular tracking of API usage. The key finding is that model popularity (by token volume) and model cost are not aligned. Cheap, open-source models see high usage volume, but expensive, closed-source models generate the majority of the expenditure.
Key Data Points for Cost Optimization
1. **Volume Leader:** DeepSeek V4 Flash is ranked #1 by token volume, indicating high adoption of this open-source model. 2. **Cost Leaders:** The cost leaderboard is dominated by Claude models, with Claude's Opus 4.8 alone accounting for approximately 25% of total spend. DeepSeek does not appear in the top 10 by cost. 3. **Implication:** High-volume, low-cost models are being used extensively, but strategic, lower-volume use of expensive frontier models accounts for most of the budget.
Actionable Optimization Considerations
Review your team's usage patterns against this industry trend. Consider: **Task Routing:** Use cheap, high-volume models for routine tasks (e.g., draft generation, summarization). Reserve expensive frontier models for tasks requiring highest capability. **Spend Monitoring:** Implement gateway-level monitoring (like Vercel's) to identify your own cost outliers. A model with low volume could still be a major cost center. **Model Evaluation:** Regularly evaluate new open-source models. The dominance of DeepSeek suggests they can handle significant workloads at a fraction of the cost, potentially allowing you to shift volume away from more expensive APIs.