/ Models Index / DeepSeek R1

DeepSeek R1

Reasoning · DeepSeek · DeepSeek · Released Jan 2025
A-

Excellent (A-) — Capability Grade

DeepSeek R1 was the most significant open-weights reasoning model release. Comparable to o1 and o3-mini at materially lower inference cost. MIT license enables unrestricted commercial deployment. Triggered material industry repricing of frontier model economics on release.

89
Composite / 100
/ Subscore Breakdown · 6 Capability Dimensions

Where this grade comes from.

General Reasoning
A-
Code Generation
A-
Math & STEM
A
Tool Use & Agency
B+
Multimodal
D
Safety & Alignment
B
/ Key Events & Disclosures

Release timeline & positioning.

  • Released Jan 2025
  • Triggered industry-wide repricing
  • MIT-licensed open weights
  • Comparable to o1 / o3-mini on benchmarks

/ Best for

Cost-sensitive reasoning workloads; sovereign-AI deployments requiring open weights with reasoning capability.

/ Watch out for

Not multimodal. Inference compute for 671B-parameter MoE is non-trivial — the 'free weights' are not the same as low-cost inference.