← Back to Directory
DeepSeek
LLM2025-01-20

DeepSeek R1

DeepSeek R1 is an open-source reasoning model that rivals OpenAI's o1 series.

#reasoning#open-source#cheap

Overview

DeepSeek R1 is a landmark open-source reasoning model that uses reinforcement learning to achieve capabilities similar to OpenAI's o1. It is released under a permissive MIT license.

Unique Factor

Frontier reasoning capabilities in a fully open-source (MIT) model.

Key Capabilities

●
Open reasoning
●
MIT license
●
MoE architecture

Benchmarks

MMLU Score
88.5%
HumanEval (Coding)
94.3%
GPQA Diamond
83%
MATH Benchmark
91.5%

Top Use Cases

Complex Math

Solving competition-level math problems.

Example: “Solve this IMO problem: [Problem text]”

Technical Specs

Context128,000 tokens
Params671B (MoE)
LicenseMIT
ArchReasoning / MoE

API Pricing

$0.14 / 1M input tokens

Output: $2.19 / 1M tokens

✓ Free tier available
Access API

Developer

Chinese AI research lab delivering frontier-level open-source models.

Prompt Library

Browse Coding Prompts →

📋

Previous Version

Deepseek V3 →