◈

DeepSeek

by DeepSeek

4.4
Benchmark RatingLaunched 2023Free / open weights

DeepSeek's R1 reasoning model rivals frontier labs on math and code at a fraction of the cost — and the weights are open. A top pick for builders who want capability without vendor lock-in.

Open weightsReasoningLow-cost API
Visit Site

Pricing Tiers

Click a tier to see full details

Free — $0

Subscribe

Open weights, self-host

Plan: FreePrice: $0Tool: DeepSeek

Capabilities & Specs

Best for
Open reasoning & code
Context window
164K
Image generation
Voice mode
Code execution
Web browsing
Offline / open weights
Public API
Starting price
Free / open

Technical Benchmarks

Raw scores from public leaderboards (LMArena, OpenAI, Anthropic). Higher is better.

MMLU88.5%

Language Understanding

HumanEval82.6%

Code Generation

SWE-bench38.8%

Software Engineering

GSM8K97.3%

Math Reasoning

GPQA51.1%

Graduate QA

IFEval78.3%

Instruction Following

LMArena ELO1257

Chatbot Arena

Pros

  • Open weights, self-hostable
  • Extremely cheap API
  • Strong math & code reasoning

Cons

  • No first-party app/UX
  • Smaller tool ecosystem

Available Models

ModelContextPriceNote
DeepSeek V4 Pro-Max256KFree / openLatest flagship (Jul 2026)
DeepSeek V4 Pro256KFree / openFrontier open model
DeepSeek V4 Flash128KFreeFast + cheap
DeepSeek R1128KFree / openReasoning model

Tasks DeepSeek Covers

CodingData