Daily AI Model Rankings — September 29, 2026
Google’s Gemini 3.7 Flash High just muscled its way into the Vision top 10 with an ELO of 1295 — and it didn’t just sneak in, it landed at the top of the entire leaderboard, dethroning Gemini 3 Pro’s months-long reign as the best vision model available.
What Changed Today
- Vision: gemini-3.7-flash-high is new in the top 10 with an ELO of 1295, surpassing the previous leader Gemini 3 Pro (ELO: 1288) by 7 points.
Vision: A Flash Model Takes the Crown
This is a significant moment for the Vision leaderboard. Gemini 3.7 Flash High now sits at ELO 1295, making it the new #1 vision model — and it’s a Flash-tier model, not a Pro. For months, Gemini 3 Pro held the top spot at 1288, and before that, the leaderboard was dominated by full-size frontier models. A Flash variant topping the chart signals that Google has closed the quality gap between its fast/cheap inference tier and its premium tier, at least for vision tasks.
For developers, the practical implications here are potentially massive. Flash models in the Gemini family have historically been priced dramatically lower than their Pro counterparts — Gemini 3 Flash runs at roughly ~$0.15/$0.60 per MTok compared to Gemini 3 Pro’s $2/$12 per MTok. If Gemini 3.7 Flash High lands anywhere near Flash-tier pricing, you could be looking at the best vision model available at a fraction of the cost. The API model ID to watch for is likely gemini-3.7-flash-high — though keep in mind pricing and availability details haven’t been fully confirmed yet. Check the Google AI developer portal for the latest access information.
FREE GUIDE
Stop Writing Design Specs by Hand
Get the free visual guide: how AI tools generate GAMP 5 documentation directly from your PLC and DCS exports. Used by Life Sciences engineers who are done doing it manually.
No spam. Unsubscribe anytime.
It’s also worth noting the “high” suffix. Google has been using quality-tier suffixes (similar to OpenAI’s approach with gpt-5.1-high) to let developers trade latency for output quality within the same model family. The “high” variant likely runs with more compute per request, which could narrow the cost advantage — but even so, this is a Flash backbone, not a Pro one.
Current Leaders at a Glance
| Category | #1 Model | Provider | Score |
|---|---|---|---|
| Vision | gemini-3.7-flash-high 🆕 | ELO 1295 |
So What?
If you’re running vision workloads in production — image understanding, document parsing, visual QA, multimodal RAG — today’s update is a clear signal to benchmark Gemini 3.7 Flash High against whatever you’re currently using. A Flash-tier model beating every Pro and frontier model on the vision leaderboard means you may be able to get better results at lower cost and lower latency simultaneously, which almost never happens. Don’t wait for a formal announcement to start testing: spin up gemini-3.7-flash-high in your pipeline, run it against your eval set, and compare both quality and per-token cost against your current model. If the pricing lands anywhere near historical Flash rates, this could be one of those rare moments where upgrading quality and cutting costs happen in the same API swap. Move fast — the builders who re-benchmark first will ship the cost savings first.
