Artificial intelligence · 2024

o3 and ARC-AGI

A reasoning model scores 87.5% on ARC-AGI, a test built to resist memorisation.

OpenAI announced o3 on December 20, 2024; a high-compute configuration reached 87.5% on the ARC-AGI-1 semi-private set and 25% on FrontierMath, where prior models scored about 2%. It was released in April 2025 with native tool use inside its reasoning.

See it in the living tree Wikipedia

Arcs above join it to what it stood on, arcs below to what it led to. The bar places this stretch on the tree's whole timeline.

What it built on 2

Reasoning models (o1) 2024
Test-time compute scaling 2024

What it led to 3

GPT-5 2025
gpt-oss open-weight models 2025
Autonomous AI researchers ~2029

Related measurements

The dashed line marks when it first worked.

Best AI score on GPQA Diamond Latest: 95.8 % correct (2026) Source: Epoch AI, Capabilities & benchmarking (benchmark_data.zip, gpqa_diamond.csv); Anthropic: Claude 3 public launch (March 4, 2024)

Training compute of notable AI models Source: Epoch AI, Notable AI Models (CC BY 4.0)