Compare
GPT-5 Mini vs DeepSeek V4
OpenAI vs DeepSeek. Specs, benchmarks, and real per-task cost — all in one page.
Verdict
DeepSeek V4 leads on LMSYS ELO (1334 vs 1288). DeepSeek V4 is ~5.0× cheaper on a 3:1 input:output blend. DeepSeek V4 is open-weights, the other is proprietary.
OpenAI
GPT-5 MiniGPT-5’s smaller, cheaper sibling. Optimized for chat and lightweight agent tasks.
Very low latencyCheap enough for mass consumer apps
DeepSeek
DeepSeek V4The shock-the-market moment of 2026. Frontier coding quality at 1/50th the price of Opus.
Unbeatable $/qualityOpen weightsGreat at code
Pricing
Input / 1M$0.50$0.14
Output / 1M$2.00$0.28
Context256K128K
Max output16K16K
LicenseProprietaryOpen weights
Released2026-01-212026-02-20
Benchmarks
LMSYS ELO1288.01334.0
MMLU Pro78.585.4
HumanEval85.591.5
SWE-Bench—54.1
MATH79.888.7
IFEval85.485.8
Per-task cost
Summarize a 1-hour meeting transcript$0.0085$0.0022
Review a 500-line pull request$0.0080$0.0017
Answer a customer support ticket$0.0031$0.0006
Extract structured data from a resume$0.0045$0.0010
Debug a stack trace with context$0.0070$0.0014
Per-call cost using published token counts for each task. Real-world prompts vary.