GLM-4.7
358B MIT-licensed model that adds interleaved and preserved thinking for agent loops, at 73.8% SWE-bench Verified and 41.0% Terminal Bench 2.0.
Thinks before every response and tool call, retains reasoning blocks across turns (preserved thinking) and offers per-turn thinking control. Gains of +5.8 points on SWE-bench Verified and +16.5 on Terminal Bench 2.0 over GLM-4.6.
- Date
- Monday, 22 December 2025
- Lab
- Zhipu AI / Z.ai
- Kind
- open-weights
- Access
- open weights
Figures
| Measure | Value | Measured by |
|---|---|---|
| SWE-bench Verified | 73.8% +5.8 vs GLM-4.6 | company |
| Terminal Bench 2.0 | 41.0% +16.5 | company |
| HLE (with tools) | 42.8% | company |
| tau2-Bench | 87.4% | company |
Benchmarks are Zhipu-reported. Release date from Z.ai release notes and HF repo creation (2025-12-22).
Sources
This record was checked against its sources on 6 October 2026. How we check