AI.info

Benchmarks

Every benchmark the tracker carries, each a leaderboard of the models its measurer ran, with the measurer’s own names, settings and days, under its licence.

In inglese

• Text Arena — Arena, 98 models; highest: Gemini 4 Argon, 1,524
• Code Arena (WebDev) — Arena, 54 models; highest: Claude Opus 5.5, 1,806
• Vision Arena — Arena, 66 models; highest: Claude Fable 5, 1,308
• Epoch Capabilities Index — Epoch AI, 118 models; highest: Claude Opus 5.5, 166.8
• GPQA Diamond — Epoch AI, 106 models; highest: Gemini 4 Argon, 96.4%
• FrontierMath, tiers 1–3 — Epoch AI, 63 models; highest: GPT-6.1 Sol, 93.7%
• FrontierMath, tier 4 — Epoch AI, 52 models; highest: GPT-6.1 Sol, 100.0%
• SWE-bench Verified — Epoch AI, 23 models; highest: Claude Opus 4.7, 83.5%
• OTIS Mock AIME 2024–2025 — Epoch AI, 104 models; highest: Gemini 4 Argon, 100.0%
• SimpleQA Verified — Epoch AI, 56 models; highest: Gemini 4 Argon, 76.0%