Intel / article

Deepseek Is Back Among The Leading Open Weights Models With V4 Pro And V4 Flash

Source: artificialanalysis.ai
Author: Artificial Analysis articles
Date: Published Apr 24, 2026

Why it matters

V4 Pro leads open models on GDPval-AA, but its 94% rate means it nearly always answers even when wrong, which matters for reliability-sensitive agents.

Key takeaways · AI-distilled

V4 is DeepSeek's first new architecture since V3 and its first two-tier lineup. V4 Flash is 284B total / 13B active; both are hybrid thinking/non-thinking, text-only, MIT-licensed, with a 1M- , 8x V3.2's.
Pricing is $1.74/$3.48 per 1M input/output tokens for V4 Pro and $0.14/$0.28 for V4 Flash. Running the Intelligence Index cost $1,071 for Pro, versus $4,811 for Claude Opus 4.7 and $948 for Kimi K2.6.
Token use is high: V4 Pro used 190M output tokens and V4 Flash 240M to run the Index, which Artificial Analysis says explains Pro's relatively high total cost despite low per-token prices.
On AA-Omniscience, V4 Pro scored -10, an 11-point gain over V3.2 driven mainly by higher accuracy, while V4 Flash scored -23, roughly level with V3.2.
V4 Pro's GDPval-AA score of 1554 beats GLM-5.1 (1535), MiniMax-M2.7 (1514) and Kimi K2.6 (1484); V4 Flash's Index score of 47 puts it at Claude Sonnet 4.6 (max) level.

Terms in this piece · Glossary

open weights — A model whose trained parameters are published for anyone to download and run — unlike API-only models you can access but never possess.
hallucination — When a model states something false with full confidence — inventing facts, citations, or APIs that don't exist.
token — The chunk of text a model reads and writes in — roughly three-quarters of a word — and the unit AI usage is billed in.
context window — The maximum amount of text a model can consider at once — its working memory for the current conversation or task.

Read the source artificialanalysis.ai