Benchmarks / Multimodal
VISTA
Visual-language understanding benchmark (Scale Labs).
maintained by Scale (Scale Labs) · 63 observations on record
Versions
1 version
| Version | Tasks |
|---|---|
| VISTA current vista | — |
Current leaderboard
VISTA · Accuracy · top 25 of 63 systems
| # | System | Accuracy % |
|---|---|---|
| 1 | Gemini 2.5 Pro Experimental (March 2025) Google | 54.6% 53.2–56.1 |
| 2 | Gemini 2.5 Pro Preview 06.05 Google DeepMind | 54.6% 54.1–55.2 |
| 3 | GPT 5.4 Pro (2026-03-05) OpenAI | 53.9% 51.9–55.9 |
| 4 | GPT 5 Pro (2025-10-06) OpenAI | 52.4% 51.3–53.5 |
| 5 | o4-mini (high) (April 2025) OpenAI | 51.8% 51.2–52.4 |
| 6 | o4-mini (medium) (April 2025) OpenAI | 51.7% 50.6–52.7 |
| 7 | o3 Pro (high) (June 2025) OpenAI | 51.6% 51.4–51.9 |
| 8 | Gemini 3 Pro Preview Google DeepMind | 51.5% 50.7–52.3 |
| 9 | GPT 5.4 (2026-03-05)max effort OpenAI | 50.9% 49.1–52.6 |
| 10 | Gemini 2.5 Pro Preview (May 06 2025) Google | 50.8% 50.2–51.4 |
| 11 | GPT 5 Mini (2025-08-07) OpenAI | 50.4% 48.1–52.7 |
| 12 | o3 (high) (April 2025) OpenAI | 50.1% 48.9–51.2 |
| 13 | GPT 5 (2025-08-07) OpenAI | 49.7% 48.7–50.7 |
| 14 | o3 (medium) (April 2025) OpenAI | 49.6% 48.9–50.3 |
| 15 | Gemini 2.5 Flash Preview (May 2025) Google | 49.1% 48.8–49.5 |
| 16 | Claude Sonnet 4.5 20250929thinking Anthropic | 48.8% 47.1–50.4 |
| 17 | Claude Opus 4.1 20250805thinking Anthropic | 48.4% 48.0–48.9 |
| 18 | Claude 3.7 Sonnet (Feb 2025)thinking Anthropic | 48.2% 47.5–48.9 |
| 19 | o1 Pro (March 2025) OpenAI | 47.3% 45.5–49.1 |
| 20 | Gemini 2.5 Flash (April 2025) Google | 47.0% 45.7–48.3 |
| 21 | Claude Opus 4thinking Anthropic | 47.0% 46.0–47.9 |
| 22 | Gemini 3.1 Flash Lite Preview Google | 46.9% 44.2–49.7 |
| 23 | GPT 5.2 (2025-12-11) OpenAI | 46.6% 44.4–48.9 |
| 24 | Claude Opus 4.5 20251101thinking Anthropic | 46.4% 46.4–46.5 |
| 25 | Claude Opus 4.6max effort Anthropic | 46.1% 45.8–46.3 |
One row per evaluated system — reasoning-effort variants rank separately.