Cogito v1 Preview - 32B
The Cogito v1 Preview is a 32.8 billion parameter instruction-tuned generative language model developed by Deep Cogito. It stands out as a hybrid reasoning model, capable of providing direct answers or engaging in self-reflection before generating a response. This model is trained using Iterated Distillation and Amplification (IDA), an alignment strategy focused on iterative self-improvement.
Key Capabilities
- Hybrid Reasoning: Can operate in a standard direct answer mode or an 'extended thinking' mode for more complex problem-solving.
- Optimized Performance: Specifically tuned for coding, STEM subjects, complex instruction following, and general helpfulness.
- Multilingual Support: Trained in over 30 languages, offering enhanced capabilities compared to size-equivalent models.
- Extended Context: Supports a substantial context length of 128k tokens.
- Tool Calling: Fully supports single, parallel, and multiple tool calls in both standard and extended thinking modes.
Good For
- Complex Problem Solving: Ideal for tasks requiring deeper reasoning, such as advanced coding challenges or intricate STEM problems, leveraging its self-reflection capabilities.
- Multilingual Applications: Suitable for applications requiring robust performance across a wide array of languages.
- Code Generation & Scripting: Excels in coding tasks, including generating bash scripts and integrating with external tools.
- Instruction Following: Designed for precise and nuanced adherence to user instructions, making it reliable for automated workflows.
- Benchmarking: Outperforms size-equivalent counterparts on industry benchmarks in both direct and reasoning modes, as detailed in the Blog Post.