DeepSeek V4 Pro

DeepSeek🇨🇳 CNFrontier· Jun 1, 2026

ReasoningTextopen weightsAPI✎ Editorial

DeepSeek's heavyweight: 1.6T total, 49B active, 1M context, FP4/FP8 mixed precision. Codeforces 3206 Elo beats GPT-5.4. MIT license, fully open weights. $0.435/$0.87 per MTok on OpenRouter.

Editorial

Context: 1M tokens
Price / 1M tok: $1.32 in · $3.96 out
Params: 1.6T MoE (49B active)
License: MIT
Modalities: text, code → text, code

4 variants

Standard	1.6T MoE (49B active)	1M		HF · OR
0813		1M		OR
0813				HF
0813 Batch		1M	$1.32/$3.96	OR

Sources

HuggingFace OpenRouter Website

Deutsch

DeepSeeks Schwergewicht: 1.6T Gesamt, 49B aktiv, 1M Kontext, FP4/FP8-Mixed-Precision. Codeforces 3206 Elo übertrifft GPT-5.4. MIT-Lizenz, komplett open-weights. Auf OpenRouter $0.435/$0.87 per MTok.

Benchmarks

Benchmark	Domain	Value	Evidence
LongBench v2vendor number	long-context	40.2 % correct	90
SimpleQAvendor number	knowledge	46.2 % correct	81
LiveCodeBenchvendor number	coding	88.8 % pass@1	80
BigCodeBenchvendor number	coding	56.8 % pass@1	66
HMMTvendor number	math	96.2 % correct	61
HumanEvalvendor number	coding	62.8 % pass@1	57
GPQA Diamondvendor number	reasoning	91.3 % correct	50
MGSMvendor number	multilingual	81.3 % correct	44
MMLUvendor number	knowledge	87.8 % correct	43

Vendor-reported numbers are marked and count with factor 0.7 in rankings. The evidence score says how much a benchmark still tells you today.

Provenance

Detected: Jun 1, 2026
First source: kuratiert
Description: Editorial
Editorially reviewed: yes