GLM 5.2, DeepSeek V4 Flash, and Kimi K2.5
Create an agent or open an existing agent in the builder, then follow the steps below. Model and feature availability may depend on your plan.
What it does
Three new open-weight models joined the model selector, available to every agent in your workspace.
Before this
An addition to the selector, and a new shape in it. The cheap end had been covered by the small variants of the frontier families — Nano, Flash Lite. These three are open-weight models that land near the top of the quality charts instead, which is a different trade to the one those variants make.
Why it matters
They land near the top of the quality charts at a fraction of the credit cost — worth testing on high-volume agents where a frontier model is overkill.
The specs
Three open-weight models, all available on every plan and all supporting Actions.
- GLM 5.2
- 1,048,576-token context, 131,072-token output, text only. Reasoning at Smartest ($$$$) or Maximum ($$$$$).
- DeepSeek V4 Flash
- 1,048,576-token context, 384,000-token output, text only. Reasoning at Smartest ($$$$) or Maximum ($$$$$).
- Kimi K2.5
- 256,000-token context, 262,144-token output, text and images. Reasoning is a simple On switch rather than a set of levels.
None of the three read images except Kimi, so an agent that handles uploaded screenshots or scanned PDFs still wants a multimodal model. For text-only work at volume, they are the cheapest credible options in the selector.