GLM 5.2 and MiniMax M3 came close to Sonnet on coding-agent tests
Tessl compared GLM 5.2, , Kimi K2.7-code, Qwen 3.7-Plus, and across about 1,000 tasks. Each task was run twice: once normally and once with a relevant skill loaded. The skills came from , and the tasks and are public in the task-evals-for-skills dataset on .
GLM 5.2 scored 91.9 overall, scored 91.4, and scored 90.8. The cost per task was $0.289 for GLM 5.2, $0.207 for , and $0.296 for . GLM 5.2 finished slightly above Sonnet while costing slightly less.
was only 0.6 points behind Sonnet and cost about 30% less. Every model improved by about 20 points when given the relevant skill, and had the biggest gain at 24.4 points. The benchmark was shared by someone who works at Tessl, the group that ran it, so the test design and task mix matter when reading the results.
Key points
- Five models were tested on about 1,000 tasks.
- GLM 5.2 scored 91.9 overall, slightly above at 90.8.
- scored close to Sonnet while costing about 30% less per task.
- Adding the relevant skill raised every model’s score by about 20 points.
- The benchmark came from Tessl, so the evaluation method should be checked before treating it as final proof.