DeepSeek V4
DeepSeek V4 is an open-weight large language model featuring a 1M-token context window that was released in late April 2026. The model has achieved significantly lower token costs compared to closed alternatives, with NVIDIA's inference optimizations reducing costs by up to 5x on the Blackwell platform, and is available through multiple inference providers including DeepInfra and Hugging Face. DeepSeek V4 incorporates architectural techniques such as KV sharing and compressed attention for improved efficiency in long-context processing.
Updated 3 August 2026
Specifications
No specifications recorded yet.
Latest developments
📈 Data to start your week
Exponential View · 1 week ago ·
19
Agentic Misalignment in Summer 2026
TLDR Dev · 2 weeks ago ·
6
How NVIDIA’s Inference Software Stack Powers the Lowest Token Cost
NVIDIA · 1 month ago ·
33
The Unbearable Cheapness of Open Weight Models
TLDR Dev · 1 month ago ·
7
Frontier post-training recipe review with Finbarr Timbers
Interconnects · 1 month ago ·
38
Latest open artifacts (#21): Open model bonanza! Gemma 4, DeepSeek V4, Kimi K2.6, MiMo 2.5, GLM-5.1 & others. On CAISI's V4 assessment.
Interconnects · 2 months ago ·
29
Recent Developments in LLM Architectures: KV Sharing, mHC, and Compressed Attention
Ahead of AI · 2 months ago ·
35
LWiAI Podcast #243 - GPT 5.5, DeepSeek V4, AI safety sabotage
Last Week in AI · 2 months ago ·
36
July 2026
Moonshot AI releases Kimi K3, a 2.8 trillion parameter open-weight model achieving frontier-class performance Open source release
June 2026
Multiple AI companies release open-source models as pricing competition intensifies Open source release
- How NVIDIA’s Inference Software Stack Powers the Lowest Token Cost
- The Unbearable Cheapness of Open Weight Models
- Frontier post-training recipe review with Finbarr Timbers
May 2026
Amazon researchers present framework for optimizing LLM architecture to improve inference speed without sacrificing accuracy Research publication
- Latest open artifacts (#21): Open model bonanza! Gemma 4, DeepSeek V4, Kimi K2.6, MiMo 2.5, GLM-5.1 & others. On CAISI's V4 assessment.
- Recent Developments in LLM Architectures: KV Sharing, mHC, and Compressed Attention
- LWiAI Podcast #243 - GPT 5.5, DeepSeek V4, AI safety sabotage
April 2026
Relationships
Products & technology
- DeepSeek develops this model · 4 sources
- NVIDIA integrated with this model · 1 source
- Hugging Face deploys this model · 1 source