GLM-5.2
- Context
- 1M
- 128K max output
- Input price
- $1.40
- per 1M tokens · Z.ai API
- Output price
- $4.40
- per 1M tokens
- Epoch Capabilities Index
- 151.8
- #37 of 71 listed
- LMArena Text
- 1,476
- #23 of 79 listed
Where to run it
4 offerings · prices as listed| Meter | Tier | Price |
|---|---|---|
| Input | Standard | $1.40 / 1M tokens |
| Output | Standard | $4.40 / 1M tokens |
| Cache read | Standard | $0.26 / 1M tokens |
| Cache write | Standard | Free / 1M tokens |
SiliconFlow (opens in a new tab) zai-org/GLM-5.2 | ||
| Input | Standard | ¥8.00 / 1M tokens |
| Output | Standard | ¥28.00 / 1M tokens |
| Cache read | Standard | ¥2.00 / 1M tokens |
| Input | Standard | ¥8.00 / 1M tokens |
| Output | Standard | ¥28.00 / 1M tokens |
| Cache read | Standard | ¥2.00 / 1M tokens |
| Input | Standard | ¥8.00 / 1M tokens |
| Output | Standard | ¥28.00 / 1M tokens |
| Cache read | Standard | ¥2.00 / 1M tokens |
Vendor-reported scores
Published by Zhipu AI, under the conditions it states| Benchmark | Setting | Score |
|---|---|---|
| Humanity's Last Exam | text-only subset, no tools | 40.5% |
| text-only subset, with tools, 300K max context | 54.7% | |
| AIME 2026 | temperature 1.0, 163,840 max generation tokens | 99.2% |
| GPQA Diamond | temperature 1.0, 163,840 max generation tokens | 91.2% |
| SWE-bench Pro | OpenHands harness, 400K context | 62.1% |
| Terminal-Bench 2.1 | Terminus-2 harness, 4h timeout, 256K context | 81% |
| Claude Code 2.1.167 harness, avg of 5 runs | 82.7% | |
| MCP-Atlas | 500-task public subset, think mode, Gemini-3.0-Pro judge | 76.8% |
Cited scores
From third-party leaderboards, not measured by us- Epoch Capabilities Index 151.8
#37 of 71 listed models · CI 149.9–153.6
Epoch AI, 'Capabilities & benchmarking'. Published online at epoch.ai. Retrieved from 'https://epoch.ai/benchmarks'. Licensed CC BY 4.0. · License · as of Sep 26, 2026 · changes we made
- LMArena Text 1,476
#23 of 79 listed models · max effort · CI 1,471–1,480 · 43,570 votes
LMArena (Arena), leaderboard-dataset on Hugging Face, licensed CC BY 4.0 · License · as of Sep 25, 2026 · changes we made
- Epoch AI · FrontierMath Tiers 1–3 59.2%
#29 of 56 listed models · max effort · ±2.96
Also listed: low effort 54.7% · reasoning off 42.5%
Epoch AI, 'Capabilities & benchmarking'. Published online at epoch.ai. Retrieved from 'https://epoch.ai/benchmarks'. Licensed CC BY 4.0. · License · as of Sep 26, 2026 · changes we made
- Epoch AI · GPQA Diamond 91.9%
#16 of 64 listed models · max effort · ±1.61
Also listed: low effort 87.9% · reasoning off 71.2%
Epoch AI, 'Capabilities & benchmarking'. Published online at epoch.ai. Retrieved from 'https://epoch.ai/benchmarks'. Licensed CC BY 4.0. · License · as of Sep 26, 2026 · changes we made
- Epoch AI · SWE-bench Verified 78.7%
#3 of 23 listed models · max effort · ±1.87
Epoch AI, 'Capabilities & benchmarking'. Published online at epoch.ai. Retrieved from 'https://epoch.ai/benchmarks'. Licensed CC BY 4.0. · License · as of Sep 26, 2026 · changes we made
- LMArena Text · Chinese 1,519
#23 of 79 listed models · max effort · CI 1,508–1,530 · 3,128 votes
LMArena (Arena), leaderboard-dataset on Hugging Face, licensed CC BY 4.0 · License · as of Sep 25, 2026 · changes we made
- LMArena Text · Coding 1,512
#35 of 79 listed models · max effort · CI 1,506–1,519 · 12,052 votes
LMArena (Arena), leaderboard-dataset on Hugging Face, licensed CC BY 4.0 · License · as of Sep 25, 2026 · changes we made