Model Details
GLM-4.5 Air
Quick answer: GLM-4.5 Air is Zhipu AI's June 2025 lightweight/cost-optimised tier scoring 70.7% LiveCodeBench, 81.4% MMLU-Pro, 76.4% BFCL-v3, and 8.12% HLE. Cheaper than GLM-4.5.
At a Glance
Where GLM-4.5 Air leads
- 70.7% LiveCodeBench — strong coding for "Air" tier
- 81.4% MMLU-Pro — good knowledge
- 76.4% BFCL-v3 — solid function calling
- Lower cost than GLM-4.5
Where it lags
- 8.12% HLE — very low frontier exam
- Below GLM-4.5 on all metrics
- Proprietary
Best for: Cost-optimised GLM-4.5 generation deployments; function-calling pipelines where GLM-4.5 is over-budget.
What GLM-4.5 Air Is
GLM-4.5 Air is the lightweight tier of Zhipu AI's June 2025 GLM-4.5 generation. "Air" in Zhipu's naming means a more cost-efficient variant with slightly reduced performance vs the base GLM-4.5.
Specifications
| Field | Value |
|---|---|
| Organization | Zhipu AI |
| License | Proprietary |
| Release date | June 2025 |
| Modality | Text only |
Pricing
Available via Zhipu AI API at lower cost than GLM-4.5.
Public Benchmark Scores
| Benchmark | Score | Source | Date |
|---|---|---|---|
| LiveCodeBench | 70.7% | Benchgen evaluation | 2025-06 |
| MMLU-Pro | 81.4% | Benchgen evaluation | 2025-06 |
| BFCL-v3 | 76.4% | Benchgen evaluation | 2025-06 |
| Humanity's Last Exam | 8.12% | Benchgen evaluation | 2025-06 |
GLM-4.5 Air vs Alternatives
| Model | LiveCodeBench | MMLU-Pro | BFCL-v3 | License |
|---|---|---|---|---|
| GLM-4.5 Air | 70.7% | 81.4% | 76.4% | Proprietary |
| GLM-4.5 | 72.9% | 84.6% | 77.8% | Proprietary |
| GLM-5.2 | — | — | — | Proprietary |
GLM-4.5 Air vs GLM-4.5: 2.2% lower LiveCodeBench, 3.2% lower MMLU-Pro at lower cost. Choose based on cost/quality tradeoff.
Related Models & Benchmarks
- Compare: GLM-4.5 · GLM-5.2
- Benchmarks: LiveCodeBench · MMLU-Pro · BFCL-v3
Frequently Asked Questions
What is GLM-4.5 Air?
Zhipu AI's June 2025 cost-optimised model scoring 70.7% LiveCodeBench, 81.4% MMLU-Pro, 76.4% BFCL-v3. Lighter tier below GLM-4.5.Specs from Zhipu AI's GLM-4.5 Air release (June 2025) and Benchgen evaluations. Last updated 2026-07-24.