Every plan and top-up includes the same 5 coding models. DeepSeek V4.1 Flash 1017 is the recommended default; you choose which model to use. The advanced models use credits faster.
- DeepSeek V4.1 Flash 1017 · The default: fast, capable coding for everyday changes, tests and agent work.
- Claude Haiku 5.5 · Fast, focused coding, tests and tool-assisted edits.
- GPT-6 Luna · Efficient reasoning for focused fixes and everyday agent tasks.
- Kimi K3.1 Pro · For demanding coding and multi-step reasoning. Uses credits faster than the everyday models.
- Gemini 4 Argon · For complex coding and deeper analysis. Uses credits faster than the everyday models.
Codium requests, including the code and tool output a task sends, go through Vector to a provider serving the model you picked. Vector routes them only to zero-data-retention endpoints that don't collect data for training. Review the privacy policy before sending code.