Segra models
Every model on Segra runs inside sealed, verified hardware. Before a model is offered here, its hardware report must show that the model itself runs inside the sealed area, so your messages are never readable by the people running the computers. Models that only pass your messages on to an outside service are not offered.
0GM 1.0
Included0GM-1.0-35B-A3BA compact, fast mixture-of-experts model: about 35 billion parameters in total, of which roughly 3 billion work on each word. That keeps answers quick and each message cheap.
Good for
- Everyday questions and conversation
- Summarizing documents, notes and meeting transcripts
- Drafting and rewriting emails and short texts
- Quick explanations and brainstorming
For long, multi-step reasoning or large coding tasks, a Pro model does noticeably better.
GLM-5.3
ProZ.ai · glm-5.3A large open-weight mixture-of-experts model (around 750 billion parameters, about 40 billion active per word), built for software engineering and agent work: planning, using tools and carrying a task through many steps.
Good for
- Writing, reviewing and debugging code, including in Segra Desktop
- Multi-step reasoning and analysis
- Long documents and long conversations
- Technical and security questions
Uses your balance roughly 10× faster than 0GM per message, so save it for work that needs it.
Pro models need a paid account. Prices follow network rates and change over time. A token is about three quarters of a word, and each message sends the whole conversation so far. You can check each model's hardware report yourself from the shield button in the chat.