GLM-4.6
zhipu/glm-4.6The open agentic specialist, engineered for tool loops, computer use and multi-step autonomy.
200K ctx · Price on request
Category
Agents
Tool use, computer use and multi-step autonomy. Models built to act: calling tools, browsing, operating software and finishing multi-step jobs. Reach for an agent model when the job is a sequence rather than a single answer. Reliability under repeated matters more than brilliance on one turn, and stamina decides how many steps a run survives.
1 model in Agents from 1 provider
R/1M tokens These models bill on input alone. Nothing is generated in return, so the output price on each card reads “—”.
Top model
Where this category stands today
Browse all agent models →
1 model in the directory
What to look for
What to look for in Agents
One model tracked today. These are the criteria that will separate the next ones.
- 01
Tool-calling reliability
One malformed tool call breaks a loop. Reliability under repeated calls matters more than single-shot brilliance.
- 02
Cost per completed task
Agents burn tokens across many steps. Evaluate cost per finished job, not per request.
- 03
Long-context stamina
Multi-step work accumulates context. Models that stay sharp at 100K+ tokens finish more tasks.