Model Details
GLM-5
Quick answer: GLM-5 is Zhipu AI's June 2025 model scoring 75.9% BrowseComp, 77.8% SWE-Bench Verified, and 43.2% CyberGym. Proprietary — strong BrowseComp and SWE-Bench for a mid-2025 model.
At a Glance
Where GLM-5 leads
- 75.9% BrowseComp — strong web research (above Muse Spark, below Kimi K2.6)
- 77.8% SWE-Bench Verified — competitive software engineering
- 43.2% CyberGym — moderate cybersecurity performance
Where it lags
- Proprietary: no open weights
- Superseded by GLM-5.2 (Jul 2026)
Best for: Zhipu AI platform BrowseComp + SWE-Bench combined workloads; Jun 2025 snapshot for research.
What GLM-5 Is
GLM-5 is Zhipu AI's June 2025 model in the GLM-5 generation — the flagship before the GLM-5.2 update. The 75.9% BrowseComp is notably higher than GLM-4.7 (52.0%), showing significant improvement.
Specifications
| Field | Value |
|---|---|
| Organization | Zhipu AI |
| License | Proprietary |
| Release date | June 2025 |
| Modality | Text only |
Pricing
Available via Zhipu AI API (bigmodel.cn).
Public Benchmark Scores
| Benchmark | Score | Source | Date |
|---|---|---|---|
| BrowseComp | 75.9% | Benchgen evaluation | 2025-06 |
| SWE-Bench Verified | 77.8% | Benchgen evaluation | 2025-06 |
| CyberGym | 43.2% | Benchgen evaluation | 2025-06 |
GLM-5 vs Alternatives
| Model | BrowseComp | SWE-Bench Verified | License |
|---|---|---|---|
| GLM-5 | 75.9% | 77.8% | Proprietary |
| GLM-5.2 | — | — | Proprietary |
| Meta Muse Spark | — | 77.4% | Proprietary |
| Kimi K2.6 | 83.2% | — | Apache 2.0 |
GLM-5 BrowseComp (75.9%) exceeds Muse Spark SWE-Bench (77.4% is comparable). For open-source BrowseComp: Kimi K2.6 (83.2%).
Related Models & Benchmarks
- Compare: GLM-5.2 · GLM-4.5 · Kimi K2.6
- Benchmarks: BrowseComp · SWE-Bench Verified · CyberGym
Frequently Asked Questions
What is GLM-5?
Zhipu AI's June 2025 model scoring 75.9% BrowseComp, 77.8% SWE-Bench Verified. Proprietary — superseded by GLM-5.2.Specs from Zhipu AI's GLM-5 release (June 2025) and Benchgen evaluations. Last updated 2026-07-24.