Compare

Qwen 3 Max vs DeepSeek V4

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

Verdict

DeepSeek V4 leads on LMSYS ELO (1334 vs 1295). DeepSeek V4 is ~16.0× cheaper on a 3:1 input:output blend.

Qwen

Alibaba’s flagship. Strongest non-English coverage of any open model; dominant in Asian markets.

MultilingualOpen weightsCompetitive coding

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$1.60$0.14

Output / 1M$6.40$0.28

Context128K128K

Max output16K16K

LicenseOpen weightsOpen weights

Released2025-11-202026-02-20

Benchmarks

LMSYS ELO1295.01334.0

MMLU Pro84.085.4

HumanEval87.291.5

SWE-Bench—54.1

MATH85.188.7

MMMU71.8—

IFEval85.585.8

Per-task cost

Summarize a 1-hour meeting transcript$0.027$0.0022

Review a 500-line pull request$0.026$0.0017

Answer a customer support ticket$0.0099$0.0006

Extract structured data from a resume$0.014$0.0010

Debug a stack trace with context$0.022$0.0014

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

Add up to four models, tweak tasks live.