Blog
Releases and field notes
The models we open, the numbers behind them and what we learn running AI in production.
Latest
Surogate 3.7 Flash
Our own model for agentic work, served from the EU.
Surogate 3.7 Flash is our own model for agent work. It reads text and images, holds 1M tokens of context, and scores 54.6 on WildClawBench, level with the best open models. You can call it today on llm.buzz.
Earlier posts
2- New models
Surogate Speech
Small open speech models for agents.
Open speech models for agents, released one language at a time. Jackrabbit turns speech into text and Amami turns text into speech. Romanian is first, and Jackrabbit gets a 5.69% word error rate on FLEURS with 116M parameters.
- New model
Surogate Rune
The best decision model you can own.
The best decision model you can own: it answers with a decision instead of prose: which option, how likely, and how it rates on your scale. It reads your text and images, on your own hardware.
Ready for enterprise-grade AI that actually ships?
Bring a real use case. We’ll show you the path to production in weeks, not quarters.