AI Agent StoreMODEL USAGE · VERCEL AI GATEWAY
Which models are
people using?
See where requests, tokens and spending go. Use the trends to shortlist models for your own tests.
Data: Oct 1, 2026–Oct 14, 2026 · Reviewed Oct 15, 2026
These figures cover Vercel AI Gateway customers. Check a model’s quality and price before choosing it for your task.
Explore model usage
Change the view to see how the picture differs.
See which models received the largest share of requests. This view uses Gateway’s text category.
Changes are percentage points: a move from 10% to 12% is up 2 points.
- GPT-6 Luna12.28%Up 1.91 pointsPreviously 10.37%
- DeepSeek V4.1 Flash10.23%Up 1.40 pointsPreviously 8.83%
- DeepSeek V4 Flash 07316.18%Up 1.93 pointsPreviously 4.25%
- GLM 5.36.09%Up 4.88 pointsPreviously 1.20%
- GLM 5.3 Flash5.61%Up 0.73 pointsPreviously 4.88%
- GPT 5.6 Luna3.86%Down 3.56 pointsPreviously 7.42%
- Gemini 3.5 Flash Lite3.70%Up 0.02 pointsPreviously 3.68%
- Gemini 3.1 Flash Lite3.41%Down 3.31 pointsPreviously 6.72%
- Gemini 3.8 Flash3.11%Down 0.67 pointsPreviously 3.78%
- Gemini 2.5 Flash Lite2.58%Up 0.09 pointsPreviously 2.49%
- Other models42.94%Down 3.43 pointsPreviously 46.37%
Other models combines entries the source does not list separately. Every share uses the full total; missing models are not counted as zero.
Use this to build a shortlist
Start with a few models that interest you. Check their benchmark results and prices, then try them on the tasks your agent will handle. Usage alone cannot tell you how reliably a model will complete those tasks.
Request share shows how often models are used. Token share depends on the amount of text processed. Spending share combines usage and prices. A model can lead one view without leading the others.
What these figures cover
This is one platform’s anonymized usage, not global market share. Text models and all models use different totals. The text view follows Gateway’s categories, which can leave out some models that choose from fixed answers. Model names can combine versions and settings, so these figures do not compare reasoning effort.
We show 14 complete days, using UTC for the start and end of each day. The publisher may cache its exports for up to 24 hours. A missing model has no separate reported share; it may be included in Other models. Changes are shown only when the same name appears on both dates. The models grouped into Other can change over time.
The export provides percentages rather than absolute requests, tokens, spending or sample sizes. Charts round shares to two decimal places; the download preserves the published precision. Usage does not affect our Agent Fit, Practical Value or Evidence Consensus ratings.
Source and downloadable data
© 2026 Vercel. AI Gateway Leaderboard Data is licensed under CC BY 4.0. Read the CC BY 4.0 license ↗
We show requests, tokens and spending as separate views and calculate changes between reported dates. The source’s totals stay intact.
Read the publisher’s methodology ↗ · Understand the source population ↗