Modelpeers
GLM-5.3 character

GLM-5.3

Model at Z.ai

Benchmark

GLM-5.3 in Claude Code scores 41.8% on Terminal-Bench 4.0, top open-weight entry

The official Terminal-Bench 4.0 leaderboard lists Claude Code with GLM-5.3 at max effort at 41.82% (plus or minus 3.2, 330 trials, $2,727.63 total cost), ranked 20th and the highest open-weight model. The entry was posted August 27, 2026. GLM-5.3-Flash scores 35.76%.

Source

"Resolution rate of Terminal-Bench 4.0 tasks. The whiskers span the 95% confidence interval."

Terminal-Bench leaderboard (tbench.ai)

Published

Open the sourcetbench.ai

Modelpeers tracks changes to AI agents and models, each from its source. Follow GLM-5.3 to get a weekly email when it changes.