Artificial intelligence · 2025

DeepSeek-R1

An open-weights reasoning model matches o1, and the recipe is published.

Released January 20, 2025 under the MIT licence, DeepSeek-R1 showed reasoning can emerge from pure reinforcement learning with verifiable rewards (R1-Zero). It matched o1 on math and code, topped app store charts, and knocked about $590 billion off Nvidia's market value in one day.

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 3

DeepSeek-V3 2024
Reasoning models (o1) 2024
Knowledge distillation 2015

What it led to 2

Qwen3 2025
DeepSeek-V4 2026

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)