Zhipu AI (Z.ai)proprietary
Released: 2026-06-12GLM-5.2
Zhipu AI fifth-generation flagship with 1M context reasoning, advanced tool use, and bilingual English/Chinese synthesis.
Intelligence Index
90.5/ 100
Calibrated multi-domain compositeThroughput Speed
72tok/sec
Output streaming throughputTime To First Token
380ms
Initial chunk server latencyLiveBench Contam-Free
77.0%
September 2026 suite scoreLiveBench Multi-Domain Evaluation Breakdown
Monthly refreshed contamination-free test setsLogical Reasoning86.8%
Coding & Repo Repair79.5%
Mathematics (AIME)90.5%
Data Analysis & Tables76.0%
Language & Comprehension72.0%
Instruction Following66.0%
Sourced Benchmark Evaluations (61)
Verified performance across authoritative benchmarks with provenance tracking.| Benchmark | Status | Score | Trust | Date | Source Type | Provenance |
|---|---|---|---|---|---|---|
| ARC-AGI-3 | Active | 57.9% | 93 | 2026-07-17 | independent | Source ↗ |
| ARC-AGI-2 | Nearing Saturation | 69.4% | 75 | 2026-07-10 | independent | Source ↗ |
| FrontierMath | Active | 25.3% | 98 | 2026-07-07 | independent | Source ↗ |
| PutnamBench | Active | 47.1% | 90 | 2026-07-06 | independent | Source ↗ |
| SWE-Lancer | Deprecated | 54.1% | 20 | 2026-07-04 | independent | Source ↗ |
| Omni-MATH | Deprecated | 67% | 10 | 2026-07-04 | independent | Source ↗ |
| Soohak | Active | 68.8% | 98 | 2026-07-02 | independent | Source ↗ |
| FrontierCode | Active | 46.1% | 98 | 2026-07-02 | independent | Source ↗ |
| Humanity's Last Exam | Active | 39.2% | 93 | 2026-06-30 | independent | Source ↗ |
| StructFlowBench | Nearing Saturation | 56.1% | 75 | 2026-06-30 | independent | Source ↗ |
| AIME | Nearing Saturation | 86% | 75 | 2026-06-30 | independent | Source ↗ |
| OlympiadBench | Active | 74.2% | 80 | 2026-06-29 | independent | Source ↗ |
| MathArena | Active | 72.4% | 99 | 2026-06-28 | independent | Source ↗ |
| FireBench | Active | 55.4% | 95 | 2026-06-28 | independent | Source ↗ |
| Terminal-Bench | Active | 55.7% | 80 | 2026-06-28 | independent | Source ↗ |
| OpenAI-MRCR | Active | 76% | 80 | 2026-06-27 | independent | Source ↗ |
| SimpleQA | Active | 47.1% | 90 | 2026-06-27 | independent | Source ↗ |
| CodeContests | Deprecated | 51.7% | 10 | 2026-06-27 | independent | Source ↗ |
| ComplexBench | Nearing Saturation | 54.1% | 80 | 2026-06-26 | independent | Source ↗ |
| SWE-bench Pro | Active | 49.3% | 90 | 2026-06-26 | independent | Source ↗ |
| LongBench-Pro | Active | 74.2% | 80 | 2026-06-26 | independent | Source ↗ |
| ZeroSCROLLS | Nearing Saturation | 72.4% | 78 | 2026-06-25 | independent | Source ↗ |
| MultiPL-E | Deprecated | 73.1% | 10 | 2026-06-25 | independent | Source ↗ |
| LiveCodeBench | Saturated | 65.2% | 39 | 2026-06-24 | independent | Source ↗ |
| AgentIF | Nearing Saturation | 44.9% | 80 | 2026-06-24 | independent | Source ↗ |
| MATH | Saturated | 88.7% | 25 | 2026-06-24 | independent | Source ↗ |
| InfiniteBench | Nearing Saturation | 76.9% | 65 | 2026-06-24 | independent | Source ↗ |
| Mercury | Deprecated | 70% | 10 | 2026-06-23 | independent | Source ↗ |
| SCROLLS | Nearing Saturation | 72.4% | 65 | 2026-06-23 | independent | Source ↗ |
| MMLU-Pro | Nearing Saturation | 83.3% | 75 | 2026-06-23 | independent | Source ↗ |
| RULER | Nearing Saturation | 83.3% | 84 | 2026-06-22 | independent | Source ↗ |
| TabMWP | Saturated | 83.3% | 25 | 2026-06-22 | independent | Source ↗ |
| CyberSecEval | Deprecated | 71.6% | 10 | 2026-06-22 | independent | Source ↗ |
| SWE-bench Verified | Saturated | 62% | 25 | 2026-06-21 | independent | Source ↗ |
| MMLU-Redux | Saturated | 88.7% | 25 | 2026-06-21 | independent | Source ↗ |
| QuALITY | Nearing Saturation | 79.6% | 78 | 2026-06-21 | independent | Source ↗ |
| LongBench | Nearing Saturation | 77.8% | 65 | 2026-06-20 | independent | Source ↗ |
| SWE-bench | Deprecated | 54.1% | 10 | 2026-06-20 | independent | Source ↗ |
| MGSM | Nearing Saturation | 86% | 65 | 2026-06-20 | independent | Source ↗ |
| AGIEval | Saturated | 76.5% | 25 | 2026-06-19 | independent | Source ↗ |
| QASPER | Nearing Saturation | 75.1% | 65 | 2026-06-19 | independent | Source ↗ |
| TruthfulQA | Saturated | 79.6% | 25 | 2026-06-19 | independent | Source ↗ |
| GSM-Hard | Deprecated | 86.9% | 10 | 2026-06-18 | independent | Source ↗ |
| SuperGLUE | Saturated | 90.8% | 25 | 2026-06-18 | independent | Source ↗ |
| CodeXGLUE | Deprecated | 74.7% | 23 | 2026-06-17 | independent | Source ↗ |
| MT-Bench | Nearing Saturation | 6.8% | 75 | 2026-06-17 | independent | Source ↗ |
| BIG-Bench Hard | Saturated | 87.2% | 25 | 2026-06-17 | independent | Source ↗ |
| WinoGrande | Saturated | 92.6% | 28 | 2026-06-16 | independent | Source ↗ |
| MBPP+ | Nearing Saturation | 81.1% | 65 | 2026-06-16 | independent | Source ↗ |
| IFEval | Deprecated | 64.7% | 25 | 2026-06-16 | independent | Source ↗ |
| GPQA | Nearing Saturation | 79.6% | 78 | 2026-06-16 | independent | Source ↗ |
| Needle-in-a-Haystack | Saturated | 98.1% | 44 | 2026-06-15 | independent | Source ↗ |
| HumanEval+ | Saturated | 85.9% | 25 | 2026-06-15 | independent | Source ↗ |
| CommonsenseQA | Saturated | 93.5% | 28 | 2026-06-15 | independent | Source ↗ |
| MBPP | Saturated | 84.3% | 25 | 2026-06-14 | independent | Source ↗ |
| GSM8K | Saturated | 97.7% | 25 | 2026-06-14 | independent | Source ↗ |
| GLUE | Saturated | 96.1% | 28 | 2026-06-14 | independent | Source ↗ |
| MMLU | Saturated | 92.3% | 25 | 2026-06-13 | independent | Source ↗ |
| HumanEval | Saturated | 90.6% | 25 | 2026-06-13 | independent | Source ↗ |
| ARC-AGI | Saturated | 83.7% | 25 | 2026-06-13 | independent | Source ↗ |
| LiveBench (Continuous) | Active | 77% | 94 | 2025-02-01 | independent | Source ↗ |
OpenAI Compatible API ExecutionAPI Ready
from openai import OpenAI
client = OpenAI()
response = client.chat.completions.create(
model="glm-5-2",
messages=[{"role": "user", "content": "Evaluate multi-step logic problem"}],
temperature=0.2
)
print(response.choices[0].message.content)Evaluation & Sourcing Notes
Scores listed for GLM-5.2 represent verified evaluations extracted from official research papers, independent evaluation suites (HELM, LMSYS, OpenCompass, LiveBench), and verified audit reports.
All benchmarks marked as saturated or deprecated reflect historical performance where the benchmark no longer provides active discriminative power.