Playground
A very large 480B parameter model optimized for code tasks. Uses incremental inference to dynamically allocate experts based on task complexity. Provides long context windows for generating and refactoring large projects. Suitable for coding assistants, automated testing, and code review.Key Features
- Massive 480B Parameter Pool: 35B active per inference enables deepest code understanding available
- Dynamic Expert Routing: Adaptive allocation: simple=10–20B, complex=30–35B parameters
- 256K Context Window: Accommodates entire large-scale projects like full React apps with dependencies
- Production-Grade Features: Generates test suites, performs code review, suggests architectural improvements