Leaked DeepSeek V4 Benchmarks Reveal a Massive 1Million Token Context Window
News Source : Geeky Gadgets
News Summary
- Leaked benchmarks for DeepSeek V4 suggest it could set new standards for AI performance.
- The model reportedly scales between 200 billion and 1 trillion parameters, using a novel MHC (Multi-Hierarchical Context) architecture.
- Its multimodal capabilities enable it to process text, images and video seamlessly, while a token context window of 1 million tokens allows it to handle highly intricate and expansive inputs.
- Enthropic has introduced significant updates to its Claude Code platform, focusing on enhanced “computer use” functionality.
Leaked benchmarks for DeepSeek V4 have sparked significant discussion, revealing a model that reportedly scales between 200 billion and 1 trillion parameters. According to the leaks, its novel MHC (M [+7020 chars]
Never miss a story from us, subscribe to our newsletter