We added two new 1M-token reasoning models within days of each other: Z-AI GLM-5.2 and Moonshot AI Kimi K3. They overlap in headlines but differ in practice. Here’s a practical guide to picking the right one.

At a glance

Pick GLM-5.2 if you…

• Work primarily with code and text
• Want a Balanced-tier option for cheaper daily driving
• Need clean, predictable code generation and refactoring
• Run agent loops that involve code execution and file edits

GLM-5.2’s text focus keeps it fast and predictable, and the Balanced tier makes it the more economical choice for high-volume work.

Pick Kimi K3 if you…

• Need to read and reason over images alongside text (screenshots, diagrams, whiteboard photos)
• Tackle knowledge-intensive questions that benefit from a very large open-weight model
• Want the extra reasoning depth of a 2.8T-parameter model for very complex prompts
• Are OK with Power-tier pricing for heavier tasks

A side-by-side example

Prompt: “Here’s a screenshot of my database schema and a SQL query that’s slow. Find the problem and rewrite it.”

• GLM-5.2 (if the screenshot is described in text) — analyzes the query, identifies missing indexes, rewrites with CTEs, and explains the plan.
• Kimi K3 — reads the screenshot directly, cross-references it with the query text, and gives the same fix plus a visual explanation of the join order from the image.

The difference shows up whenever your prompt mixes images with text. For pure text or code, both are excellent — pick by tier preference and price.

Both are live today

You don’t have to commit. Switch between GLM-5.2 and Kimi K3 in the dropdown between messages. Many Chat-O users keep one Balanced-tier model and one Power-tier model in rotation and pick per task.