VISTA
Vision-language understanding. Scale AI.
As of 25 Oct 2026, Gemini 2.5 Pro leads VISTA on BenchLeader with 54.6%, ahead of GPT-5.4 Pro at 53.9%, across 57 model configurations with a published result.
- Published by
- Scale AI SEAL
- Category
- Multimodal
- Index weight
- 1.0
- Models
- 57
- Data as of
- 25 Oct 2026
- Newest result
- 23 Mar 2026
The publisher has posted no new results on this board since 23 Mar 2026. BenchLeader re-reads it every day, so newer models appear here as soon as they are published.
Scale AI SEAL Leaderboards.
What the test looks like
VISTA measures vision-language understanding across image reasoning tasks written for the benchmark.
How it is scored
Accuracy, published by Scale AI.
What to keep in mind
Only models with vision are listed; task mix is Scale's.
- 1Gemini 2.5 Pro54.6%
- 2GPT-5.4 Pro53.9%
- 3GPT-5 Pro52.4%
- 4o4-mini (high)51.8%
- 5o4-mini (medium)51.7%
- 6o3-pro (high)51.6%
- 7Gemini 3 Pro51.5%
- 8GPT-5.4 (thinking)50.9%
- 9GPT-5 mini50.4%
- 10o3 (high)50.1%
- 11GPT-549.7%
- 12o3 (medium)49.6%
- 13Gemini 2.5 Flash49.1%
- 14Claude Sonnet 4.5 (thinking)48.8%
- 15Claude Opus 4.1 (thinking)48.4%
57 of 57
| # | ||||
|---|---|---|---|---|
| 1 |  Gemini 2.5 Pro | 54.6% | 53.9 | 2025-03-05 |
| 2 |  GPT-5.4 Pro | 53.9% | 63.3 | 2026-03-23 |
| 3 | GPT-5 Pro | 52.4% | 59.8 | 2025-11-06 |
| 4 | o4-minihigh | 51.8% | 50.1 | 2025-04-25 |
| 5 | o4-minimedium | 51.7% | 51.9 | 2025-04-16 |
| 6 | o3-prohigh | 51.6% | 53.7 | 2025-06-26 |
| 7 | Gemini 3 Pro | 51.5% | 60.4 | 2025-11-19 |
| 8 | GPT-5.4thinking | 50.9% | – | 2026-03-10 |
| 9 | GPT-5 mini | 50.4% | 56.5 | 2025-08-22 |
| 10 | o3high | 50.1% | 53.9 | 2025-04-25 |
| 11 | GPT-5 | 49.7% | 56.2 | 2025-08-13 |
| 12 | o3medium | 49.6% | 53.3 | 2025-04-16 |
| 13 | Gemini 2.5 Flash | 49.1% | 52.7 | 2025-05-20 |
| 14 |  Claude Sonnet 4.5thinking | 48.8% | 52.6 | 2025-10-02 |
| 15 | Claude Opus 4.1thinking | 48.4% | 56.4 | 2025-08-08 |
| 16 | Claude 3.7 Sonnetthinking | 48.2% | 51.0 | 2025-03-05 |
| 17 | o1-pro | 47.3% | – | 2025-03-05 |
| 18 | Claude Opus 4thinking | 47.0% | 55.0 | 2025-05-24 |
| 19 | Gemini 3.1 Flash Lite | 46.9% | 53.7 | 2026-03-23 |
| 20 | GPT-5.2 | 46.6% | 58.0 | 2025-12-15 |
| 21 | Claude Opus 4.5thinking | 46.4% | 59.9 | 2025-11-26 |
| 22 | Claude Opus 4.6max | 46.1% | 57.4 | 2026-02-17 |
| 23 | Gemini 2.0 Flash Thinkingthinking | 45.5% | 44.9 | 2025-03-05 |
| 24 | Claude Sonnet 4thinking | 45.5% | 53.2 | 2025-05-23 |
| 25 | Claude Opus 4.6no reasoning | 45.5% | 56.5 | 2026-02-17 |
| 26 | GPT-4.1 | 45.3% | 48.3 | 2025-04-15 |
| 27 | Claude Opus 4.5 | 45.3% | 57.9 | 2025-11-26 |
| 28 | o1 | 45.3% | 53.1 | 2025-03-05 |
| 29 | Claude Opus 4.1 | 45.3% | 51.5 | 2025-08-08 |
| 30 | Claude Sonnet 4.5 | 45.0% | 53.7 | 2025-10-02 |
| 31 | GPT-5.1thinking | 43.8% | 54.1 | 2025-11-26 |
| 32 | Claude Opus 4 | 43.5% | 52.6 | 2025-05-23 |
| 33 | Gemini 2.0 Pro | 43.3% | 47.0 | 2025-03-05 |
| 34 | Claude Sonnet 4 | 43.2% | 50.5 | 2025-05-23 |
| 35 | Claude 3.7 Sonnet | 43.0% | 50.0 | 2025-03-05 |
| 36 | GPT-4.5 | 42.1% | 49.7 | 2025-03-05 |
| 37 |  Kimi K2.5open ↗ | 41.9% | 54.5 | 2026-02-12 |
| 38 | GPT-4.1 mini | 41.1% | 45.0 | 2025-04-14 |
| 39 | Gemini 2.0 Flash | 40.0% | 43.2 | 2025-03-05 |
| 40 | Claude 3.5 Sonnet | 38.7% | 46.3 | 2025-03-05 |
| 41 |  Llama 4 Maverickopen ↗ | 38.3% | 42.8 | 2025-04-07 |
| 42 | GPT-4o | 38.0% | 43.9 | 2025-03-05 |
| 43 | Gemini 1.5 Pro | 37.1% | 46.1 | 2025-03-05 |
| 44 | GPT 5.1 Instant | 34.9% | – | 2025-11-26 |
| 45 |  Mistral Medium 3 | 34.6% | 44.1 | 2025-05-13 |
| 46 | Gemini 1.5 Flash 002 | 34.0% | 38.3 | 2025-03-05 |
| 47 | Pixtral Largeopen | 33.9% | 41.9 | 2025-03-05 |
| 48 | Gemini 2.0 Flash-Lite | 32.7% | 47.2 | 2025-03-05 |
| 49 |  Qwen2-VL-72B | 28.6% | – | 2025-03-05 |
| 50 | Claude 3 Opus | 27.8% | 39.2 | 2025-03-05 |
| 51 | GPT-4.1 nano | 26.6% | 38.0 | 2025-04-14 |
| 52 |  Nova Pro | 26.3% | 39.8 | 2025-03-05 |
| 53 | Pixtral 12Bopen | 26.0% | – | 2025-03-05 |
| 54 | Nova Lite | 25.5% | 38.3 | 2025-03-07 |
| 55 | Llama 3.2 90B Visionopen ↗ | 24.6% | 36.9 | 2025-03-07 |
| 56 | Llama 3.2 Instruct 11B (Vision)open ↗ | 20.5% | 36.5 | 2025-03-27 |
| 57 |  Phi 3.5 Vision | 15.2% | – | 2025-03-27 |
More multimodal benchmarks · All LLM benchmarks
Cite as: BenchLeader, “VISTA leaderboard”, data as of 25 Oct 2026, https://www.benchleader.com/benchmarks/scale_vista.
VISTA: questions
- What does VISTA measure?
- VISTA measures vision-language understanding across image reasoning tasks written for the benchmark. Scores are reported in percent of tasks solved; higher is better.
- Which AI model leads VISTA?
- Gemini 2.5 Pro leads VISTA with 54.6% as of 25 Oct 2026, ahead of GPT-5.4 Pro at 53.9%.
- How many models have VISTA results?
- 57 model configurations have a VISTA result on BenchLeader, all taken from Scale AI SEAL.
- Who runs VISTA and how often is it updated?
- VISTA is published by Scale AI SEAL. BenchLeader re-reads the published results once a day and records the date each result was published.
- Does VISTA count toward the BenchLeader Index?
- Yes. VISTA contributes to the multimodal category of the BenchLeader Index, normalised so that 50 is the average of the evaluated models.