Compare
DeepSeek V4 vs GPT-5 Turbo
DeepSeek vs OpenAI. Specs, benchmarks, and real per-task cost — all in one page.
Verdict
GPT-5 Turbo leads on LMSYS ELO (1398 vs 1334). DeepSeek V4 is ~60.0× cheaper on a 3:1 input:output blend. DeepSeek V4 is open-weights, the other is proprietary.
DeepSeek
DeepSeek V4The shock-the-market moment of 2026. Frontier coding quality at 1/50th the price of Opus.
Unbeatable $/qualityOpen weightsGreat at code
OpenAI
GPT-5 TurboOpenAI’s flagship unified model. Handles text, vision, and audio natively. The generalist benchmark champion.
Native multimodalStrong math and scienceHuge ecosystem
Pricing
Input / 1M$0.14$6.00
Output / 1M$0.28$24.00
Context128K400K
Max output16K32K
LicenseOpen weightsProprietary
Released2026-02-202026-01-21
Benchmarks
LMSYS ELO1334.01398.0
MMLU Pro85.491.2
HumanEval91.593.0
SWE-Bench54.165.8
MATH88.791.5
GPQA—68.1
MMMU—78.8
IFEval85.888.9
Per-task cost
Summarize a 1-hour meeting transcript$0.0022$0.102
Review a 500-line pull request$0.0017$0.096
Answer a customer support ticket$0.0006$0.037
Extract structured data from a resume$0.0010$0.054
Debug a stack trace with context$0.0014$0.084
Per-call cost using published token counts for each task. Real-world prompts vary.