DeepSeek V4
By DeepSeek
DeepSeek's current generation, scored as V4 Pro by Artificial Analysis at 44 on the Intelligence Index and 47 on the Coding Agent Index, with the cheapest cost per task in the scored set ($0.04) at $0.435/$0.87 per million tokens and a 1M-token context.
Visit DeepSeek V4AI-Powered
Leverages advanced AI technology to deliver cutting-edge capabilities and results.
Fast & Efficient
Optimized performance ensures quick results without compromising on quality.
Purpose-Built
Specifically designed for llms tasks and workflows.
Development Timeline
Architecture Unveiled
DeepSeek revealed the V4 architecture featuring "Engram Memory," a persistent storage layer that survives across sessions.
Open Source Milestone
The trillion-parameter model weights were released under a permissive license, disrupting the proprietary model landscape.
Frequently Asked Questions
How does DeepSeek V4 "Engram Memory" work?
Engram Memory is a specialized neural cache that stores intermediate reasoning steps and long-term facts in a way that remains accessible across different chat sessions, significantly reducing the needle-in-a-haystack retrieval problem.
Is DeepSeek V4 really open source?
Yes, DeepSeek continues its commitment to open weights by releasing the full configuration and weights of the V4 model, allowing for local deployment on enterprise-grade GPU clusters.
Specifications
AI Evaluation
The value outlier: an order of magnitude cheaper per task than the Western frontier, at a meaningful but not disqualifying capability gap.
Pros
- Cheapest cost per task in the scored set ($0.04)
- 1M-token context
- Open-weights heritage
Cons
- Intelligence Index (44) well behind the frontier tier
- Serving capacity has historically been constrained at peak times
Related Tools
FLUX
FLUX, from Black Forest Labs, is a family of high-quality open and commercial image generation models prized for photorealism and prompt adherence. Widely integrated across third-party tools and APIs, it has become a default backbone for image generation.
Claude Opus 5.5
Claude Opus 5.5, launched 22 September 2026, is Anthropic's current Opus-line flagship, replacing Opus 5. It performs at roughly the level of Claude Fable 5.1 on most work while costing 40% less to run than Opus 5, at $4 input / $20 output per million tokens (both 20% down, cache reads cut 60% to $0.20). It leads on Terminal-Bench 4.0, GDPval-AA v2.1 and Humanity's Last Exam, and is the first Opus model to ship with Fable-5.1-class cybersecurity, biology and distillation safeguards.
Claude Sonnet 5.5
Claude Sonnet 5.5, released 28 September 2026, is the second model in Anthropic's Claude 5.5 family and the faster, lower-cost complement to Claude Opus 5.5. It keeps Sonnet 5's $2 input / $10 output per million tokens ($0.20 cache reads) while generating output over 30% faster, and scores 70.6% on Terminal-Bench 4.0, above Opus 5.5's 66.4%. It has a 1M-token context window, 128K max output, adaptive thinking, and is the first Sonnet with cyber and anti-distillation classifiers.