GLM-4.5-Air
Announced on z.ai
Benchmark results
| Benchmark | Score | Source check |
|---|---|---|
| SWE-bench VerifiedOpenHands v0.34.0; up to 100 iterations; 128K context with history truncation | 57.6% | |
| AIME 2024No tools · thinking mode; average over 32 samples, answers checked by an LLM | 89.4% | |
| Terminal-Bench 1.0Terminus harness, standard function calling | 30.0% | |
| MMLU-Prothinking mode | 81.4% | |
| LiveCodeBenchthinking mode; problems from Jul 2024 to Jan 2025 | 70.7% |