GPT-4o mini is the stronger all-rounder.
Scores updated · 20 tests both models report · How we compare
Where each one wins
Tests won in each of the one area where both have results. Each piece is one test, so a longer bar means more evidence; grey means the two scored within a point of each other.
What it costs
Prices per million tokens, roughly 750,000 words. The bars show the cost of a million tokens read plus a million written.
The biggest differences
The tests each model wins by the widest margin, up to three each. Scores are out of 100.
Every test, side by side
All 20 tests both models report. The winning score is in its model's colour; marks a score checked independently.
Images and chartsGPT-4o mini
Other results17 tests, not counted
Tests outside the eight areas. They are not counted above: several are summary scores built from other tests, or the same test under another name.
- ScreenSpot-V2Kimi VL A3B Instruct by 85.96.992.8+85.9
- MLVU-MCQ (Acc)Kimi VL A3B Instruct by 26.148.174.2+26.1
- InfoVQA (Acc)Kimi VL A3B Instruct by 25.357.983.2+25.3
- MMVU-Val (Pass@1)GPT-4o mini by 9.461.652.2+9.4
- OCRBench (Acc)Kimi VL A3B Instruct by 82 rating points785867+82 rating
- WindowsAgentArena (Pass@1)Kimi VL A3B Instruct by 7.72.710.4+7.7
- AI2D (Acc)Kimi VL A3B Instruct by 7.177.884.9+7.1
- MMStar (Acc)Kimi VL A3B Instruct by 6.554.861.3+6.5
- LongVideoBench (val)Kimi VL A3B Instruct by 6.358.264.5+6.3
- MMLongBench-DOC (Acc)Kimi VL A3B Instruct by 6.12935.1+6.1
- MMBench-EN-v1.1 (Acc)Kimi VL A3B Instruct by 677.183.1+6
- BLINK (Acc)Kimi VL A3B Instruct by 3.753.657.3+3.7
- VideoMME (w sub.)Kimi VL A3B Instruct by 3.768.972.6+3.7
- Video-MME (w/o sub.)Kimi VL A3B Instruct by 364.867.8+3
- TOMATOKimi VL A3B Instruct by 2.928.831.7+2.9
- RealWorldQA (Acc)Kimi VL A3B Instruct by 167.168.1+1
- MMVet (Pass@1)tie66.966.7tie
Questions people ask
Which is better, GPT-4o mini or Kimi VL A3B Instruct?
GPT-4o mini wins the one area where both have results: images and charts. Kimi VL A3B Instruct wins none.
How do you compare the two?
We use the 20 benchmark tests both models have published scores on. The verdict counts the 3 tests in the eight capability areas, and a gap under one point (ten on rating-style scales) counts as a tie. The other 17 are listed but not counted, because several are summary scores or repeat a test. Each score is the one shown on the model's own page.