Compare

DeepSeek R2 vs DeepSeek V4

DeepSeek vs DeepSeek. Specs, benchmarks, and real per-task cost — all in one page.

Verdict

DeepSeek V4 is ~5.5× cheaper on a 3:1 input:output blend.

DeepSeek

DeepSeek’s reasoning variant. Competes with o4 on math at a fraction of the cost.

Frontier math at open-weights pricingLong CoT traces

DeepSeek

The shock-the-market moment of 2026. Frontier coding quality at 1/50th the price of Opus.

Unbeatable $/qualityOpen weightsGreat at code

Pricing

Input / 1M$0.55$0.14

Output / 1M$2.19$0.28

Context128K128K

Max output64K16K

LicenseOpen weightsOpen weights

Released2026-01-152026-02-20

Benchmarks

LMSYS ELO—1334.0

MMLU Pro87.885.4

HumanEval90.191.5

SWE-Bench—54.1

MATH94.188.7

GPQA71.2—

IFEval—85.8

Per-task cost

Summarize a 1-hour meeting transcript$0.0093$0.0022

Review a 500-line pull request$0.0088$0.0017

Answer a customer support ticket$0.0034$0.0006

Extract structured data from a resume$0.0049$0.0010

Debug a stack trace with context$0.0077$0.0014

Per-call cost using published token counts for each task. Real-world prompts vary.

Add up to four models, tweak tasks live.