Z.ai
Z.ai

GLM-5

Release Date 2026-02-11(Knowledge cutoff 2026-02-17)80K IN · 131K OUTOpen Model(MIT)API available

GLM-5 is Z.ai's MIT-licensed mixture-of-experts language model for complex systems engineering and sustained tool workflows. The developer describes expanded model scale and training data, sparse attention, and reinforcement learning, with official weights and API access offering different deployment paths. AIB currently places coding, automation, and document work in the middle, so launch-time leadership claims should not be generalized to every present-day use. Consider it for code generation, debugging, and task planning, while the surrounding application executes tool requests and checks whether they succeeded. Self-hosting must account for the full weight footprint and long-input costs; evaluate iterative tasks by final accuracy and time to completion.

API cost

Market price (median)
$0.95IN$2.88OUT
Price range7 providers
$0.6~$1IN$1.92~$3.2OUT
IN
OUT
10/2311/22
SourceOpenRouterLast updated

Output Speed

Very slowper 2,000 tokens
43tok/s
First Output 1.8stotal: 48.3s
SourceOpenRouterLast updated

User Comparisons

Response Speed
First response 23.1sTotal 44.3s
49 tok/s
SourceAIB Inc.Last updated

Expert Performance Evaluation

145727.9

More about Arena Text

$4,43298.2%67.8%82.5%73.7%

More about Vending-Bench 2

75.7%145262.9%53.2%18.7%

By effortNone 43.7%Thinking 75.7%

More about Long Context Reasoning

145262.0%50.4%10.1%89.9%60.2

More about Arena Text Factuality

149876655.1%43.2%44.2143546.2%72.1%48.2%148823.4%52.4%

More about Arena Coding

1447147172.3%144714471438

More about Arena Writing, Literature & Language

29.3%144382.0%80.0%82.5%1479145516.4%2.1%44.7%4.9%53.2%

By effortNone 7.6%Thinking 29.3%

More about Humanity's Last Exam

Specialized fields

14241415147814811479146843.2%96.7%99.2%93.1%89.0%97.5%86.4%94.2%74.3%77.0%71.7%86.6%89.7%95.8%84.2%91.7%73.0%88.5%43.8%26.8%74.4%88.0%89.0%82.0%72.7%87.9%61.6%10.9%68.6%78.8%84.2%77.1%91.0%35.1%

More about Arena Korean

OddBench

28.0%Reasoning effort: High74.8%31

By effortNone 20.0%High 28.0%

More about BullshitBench v2

See each evaluation’s detail page for its source and last updateData corrections & removal requests