GLM-5.2
- Made by
- Z.ai
- Released
- Announced in
- GLM-5.2: Built for Long-Horizon Tasksz.ai
- Event
- GLM-5.2
- API record
- /api/v1/models/glm-5-2
Benchmark results
| Benchmark | Score | Source check |
|---|---|---|
| Humanity's Last ExamNo tools · text-only subset; GPT-5.5 judge; up to 163,840 output tokens | 40.5% | CheckedA person checked this number against its source on 19 Oct 2026. This confirms the number matches what the lab published, not that the score was reproduced. |
| Humanity's Last ExamWith tools · text-only subset; 300,000-token context, no context management | 54.7% | CheckedA person checked this number against its source on 19 Oct 2026. This confirms the number matches what the lab published, not that the score was reproduced. |
| SWE-bench ProOpenHands with a tailored prompt; 400K context · The source doesn't say whether this is the public or the commercial set | 62.1% | CheckedA person checked this number against its source on 19 Oct 2026. This confirms the number matches what the lab published, not that the score was reproduced.CorroboratedThe same number is printed in 1 other lab document: 1 |
| GPQA Diamondup to 163,840 output tokens | 91.2% | CheckedA person checked this number against its source on 19 Oct 2026. This confirms the number matches what the lab published, not that the score was reproduced.CorroboratedThe same number is printed in 1 other lab document: 1 |
| AutomationBenchAutomationBench v1.0.6 · reported later, in GLM-5.3: Frontier Coding with Emergent Cyber Capabilities | 26.2% | CheckedA person checked this number against its source on 19 Oct 2026. This confirms the number matches what the lab published, not that the score was reproduced. |
| DeepSWE v1.1mini-swe-agent harness; 2h timeout; 400K context | 46.2% | CheckedA person checked this number against its source on 19 Oct 2026. This confirms the number matches what the lab published, not that the score was reproduced. |
| Terminal-Bench 2.1Terminus 2 harness; 4h timeout; 256K context | 81.0% | CheckedA person checked this number against its source on 19 Oct 2026. This confirms the number matches what the lab published, not that the score was reproduced.CorroboratedThe same number is printed in 1 other lab document: 1 |
| Claude Code harness; no time limit; averaged over 5 runs | 82.7% | CheckedA person checked this number against its source on 19 Oct 2026. This confirms the number matches what the lab published, not that the score was reproduced. |
| MCP Atlasmaximum thinking effort; the post's chart, where its table prints 76.8 | 77.0% | CheckedA person checked this number against its source on 19 Oct 2026. This confirms the number matches what the lab published, not that the score was reproduced. |
| 500-task public set; thinking mode; 10-minute timeout per task; Gemini 3 Pro judge | 76.8% | CheckedA person checked this number against its source on 19 Oct 2026. This confirms the number matches what the lab published, not that the score was reproduced. |
Sources
Reported by Z.ai
- GLM-5.2: Built for Long-Horizon Tasksz.ai · 16 Jun 2026
Original image 
Our capture of the source - GLM-5.3: Frontier Coding with Emergent Cyber Capabilitiesz.ai · 14 Aug 2026
Our capture of the source
