Skip to main content
We build AI models optimized for one metric: getting real work done. Not chasing benchmark scores or abstract intelligence tests — but practical, reliable performance on real-world tasks.
You don’t choose the model — our API automatically selects the best model for each request. This lets us continuously improve performance without requiring any changes on your end.

​
Agent 4

Latest Release — Our most capable model yet.
Agent 4 is our flagship agentic model, designed for complex multi-step tasks. It excels at reasoning, planning, and tool use — making it ideal for autonomous workflows. Best for: Complex tasks, multi-step reasoning, agentic applications

​
Small 2.0

A lightweight, efficient model optimized for speed without sacrificing quality. Perfect for high-throughput applications where latency matters. Best for: Fast responses, high-volume applications, cost-efficient deployments

​
Big 1.0

Our larger architecture model for demanding intelligence tasks. Developed alongside Small 2.0 as part of our model scaling research.
Internal release only.

​
SWE-LLM

Our first experimental model, fine-tuned for the Swedish language. At launch, it was the largest Swedish LLM available — trained on over 12,700 Swedish texts (13.3M+ tokens). Best for: Swedish language generation, Swedish conversational AI

Open Source on Hugging Face

SWE-LLM is open source. We believe in contributing to the community and plan to release more models in the future.