BlogLLM Benchmark Scores as Published, Side by Side
GLM-5.2: Benchmark Scores as Published
Every result the catalog holds for GLM-5.2: 69 tests in 13 areas, from 13 pages. 30 printed by Z.ai itself, the rest by the company named beside each number. Each number is as printed; nothing is ranked or rescored.
Developer Z.aiReleased 2026-06-16Context window 1MMax output 128KInput TextOutput TextWeights Open (MIT)Data version 2026103001, as of 2026-10-30
Knowledge Work
Agentic Coding
| Test | Value | Printed by | Setup as printed |
|---|---|---|---|
| SWE-Bench Pro | 62.1% | Z.ai, GLM-5.2 model cardchecked 2026-10-15; printed on 3 more pages | Harness: OpenHands; Context: 400000 |
| Terminal-Bench 2.1 | 82.7% | Z.ai, GLM-5.2 model cardchecked 2026-10-15; printed on 7 more pages | - |
| 81% | Z.ai, GLM-5.2 model cardchecked 2026-10-15 | Harness: Terminus-2; Context: 256000 | |
| DeepSWE v1.1 | 46.2% | Z.ai, GLM-5.2 model cardchecked 2026-10-15; printed on 6 more pages | Harness: mini-swe-agent; Context: 400000 |
| NL2Repo-Bench | 48.9 | Z.ai, GLM-5.2 model cardchecked 2026-10-15; printed on 3 more pages | Context: 400000 |
| ProgramBench Almost@1 | 9.5% | Z.ai, GLM-5.3 model cardchecked 2026-10-15 | - |
| SWE-Marathon | 13% | Z.ai, GLM-5.2 model cardchecked 2026-10-15; printed on 1 more page | Context: 1000000 |
| ProgramBench | 63.7% | Z.ai, GLM-5.2 model cardchecked 2026-10-15; printed on 1 more page | Harness: Claude Code 2.1.156; Context: 400000 |
| Terminal-Bench 3.0 | 4.6% | Z.ai, GLM-5.3 model cardchecked 2026-10-15 | - |
| FrontierSWE (dominance score) | 74.4 | Z.ai, GLM-5.2 model cardchecked 2026-10-15; printed on 2 more pages | Context: 1000000 |
| Kimi Code Bench 2.0 | 64.2% | Moonshot AI, Kimi K3 model cardchecked 2026-10-15 | Effort: max; Harness: Claude Code |
| SWE-Marathon v1.1 (resolution rate) | 19.4% | Z.ai, GLM-5.3 model cardchecked 2026-10-15 | - |
| DSBench-FullStack | 61.8% | DeepSeek, DeepSeek-V4-Flash-0731 model cardchecked 2026-10-15; printed on 1 more page | - |
| DSBench-Hard | 54.5% | DeepSeek, DeepSeek-V4-Flash-0731 model cardchecked 2026-10-15; printed on 1 more page | - |
| SWE-Atlas-QnA (strict) | 46.4% | IFM, IFM K2-Horizon-375B-A23B model cardchecked 2026-10-15 | Effort: max; Tools: no internet (strict) |
| Z.ai Code Bench | 23.4% | Z.ai, GLM-5.3 postchecked 2026-10-15 | Effort: Max |
ML Engineering
| Test | Value | Printed by | Setup as printed |
|---|---|---|---|
| PostTrainBench v1.1 | 34.3% | Z.ai, GLM-5.2 model cardchecked 2026-10-15; printed on 2 more pages | Context: 1000000 |
| MLS-Bench-Lite | 40.4 | Moonshot AI, Kimi K3 model cardchecked 2026-10-15 | Effort: max; Harness: Claude Code |
Science and Math
| Test | Value | Printed by | Setup as printed |
|---|---|---|---|
| GPQA Diamond | 91.2% | Z.ai, GLM-5.2 model cardchecked 2026-10-15; printed on 3 more pages | - |
| SciCode | 50.5% | IFM, IFM K2-Horizon-375B-A23B model cardchecked 2026-10-15; printed on 1 more page | Effort: max |
| IMO-AnswerBench | 91% | Z.ai, GLM-5.2 model cardchecked 2026-10-15 | - |
| AIME 2026 | 99.2% | Z.ai, GLM-5.2 model cardchecked 2026-10-15; printed on 1 more page | - |
| HMMT November 2025 | 94.4% | Z.ai, GLM-5.2 model cardchecked 2026-10-15 | - |
| HMMT February 2026 | 92.5% | Z.ai, GLM-5.2 model cardchecked 2026-10-15 | - |
| CritPt | 20.9% | Z.ai, GLM-5.2 model cardchecked 2026-10-15; printed on 2 more pages | - |
| SciCode-Verified | 88.1% | Mistral AI, Mistral Large 4 (preview)checked 2026-10-15 | Attempt budget: pass@1 (n=6) |
Long Context
| Test | Value | Printed by | Setup as printed |
|---|---|---|---|
| AA-LCR | 76.7% | IFM, IFM K2-Horizon-375B-A23B model cardchecked 2026-10-15; printed on 1 more page | Effort: max |
Cybersecurity
| Test | Value | Printed by | Setup as printed |
|---|---|---|---|
| CyberGym | 77.2% | Z.ai, GLM-5.3 model cardchecked 2026-10-15 | - |
| ExploitBench | 24.4% | Z.ai, GLM-5.3 model cardchecked 2026-10-15 | - |
| ExploitGym (2h) | 29 | Z.ai, GLM-5.3 model cardchecked 2026-10-15 | Attempt budget: timeout budget 2 hours (rescaled API inference time plus overhead) |
| ExploitGym (6h) | 39 | Z.ai, GLM-5.3 model cardchecked 2026-10-15 | Attempt budget: timeout budget 6 hours (rescaled API inference time plus overhead) |
| Cybench | 73% | Mistral AI, Mistral Large 4 (preview)checked 2026-10-15 | - |
General Knowledge and Reasoning
| Test | Value | Printed by | Setup as printed |
|---|---|---|---|
| Humanity's Last Exam (no tools) | 40.5% | Z.ai, GLM-5.2 model cardchecked 2026-10-15; printed on 3 more pages | Tools: none |
| Humanity's Last Exam (with tools) | 54.7% | Z.ai, GLM-5.2 model cardchecked 2026-10-15; printed on 4 more pages | Tools: tools; Context: 300000 |
| AA-Omniscience Accuracy | 24% | IFM, IFM K2-Horizon-375B-A23B model cardchecked 2026-10-15 | Effort: max |
| SimpleQA Verified | 38.1% | Thinking Machines, Thinking Machines Inkling model cardchecked 2026-10-15 | - |
| AA-Omniscience Index | 4 | Thinking Machines, Thinking Machines Inkling model cardchecked 2026-10-15 | - |
| AA-Omniscience Non-Hallucination | 74% | IFM, IFM K2-Horizon-375B-A23B model cardchecked 2026-10-15 | Effort: max |
Tool Use and Workflows
| Test | Value | Printed by | Setup as printed |
|---|---|---|---|
| Toolathlon | 48.2% | Z.ai, GLM-5.2 model cardchecked 2026-10-15 | - |
| τ³-bench Banking | 34.6% | IFM, IFM K2-Horizon-375B-A23B model cardchecked 2026-10-15; printed on 2 more pages | Effort: max |
| Agents' Last Exam (pass rate) | 23.8% | Z.ai, GLM-5.3 model cardchecked 2026-10-15; printed on 4 more pages | - |
| Toolathlon Verified | 59.9% | Z.ai, GLM-5.3 model cardchecked 2026-10-15; printed on 4 more pages | - |
| MCP Atlas | 82.6% | Moonshot AI, Kimi K3 model cardchecked 2026-10-15; printed on 1 more page | Effort: max |
| MCP Atlas (public set) | 76.8% | Z.ai, GLM-5.2 model cardchecked 2026-10-15 | - |
| MCPMark | 72.4% | IFM, IFM K2-Horizon-375B-A23B model cardchecked 2026-10-15 | Effort: max |
| WildClawBench | 55% | IFM, IFM K2-Horizon-375B-A23B model cardchecked 2026-10-15 | Effort: max |
Web Search and Research
| Test | Value | Printed by | Setup as printed |
|---|---|---|---|
| ResearchRubrics | 71.1% | Moonshot AI, Kimi K3 model cardchecked 2026-10-15 | Effort: max |
Instruction Following
| Test | Value | Printed by | Setup as printed |
|---|---|---|---|
| IFBench | 73.3 | Thinking Machines, Thinking Machines Inkling model cardchecked 2026-10-15 | - |
Multilingual
| Test | Value | Printed by | Setup as printed |
|---|---|---|---|
| Global-MMLU-Lite | 89.2% | Thinking Machines, Thinking Machines Inkling model cardchecked 2026-10-15 | - |
Safety and Alignment
| Test | Value | Printed by | Setup as printed |
|---|---|---|---|
| FORTRESS (Adversarial) | 71.3% | Thinking Machines, Thinking Machines Inkling model cardchecked 2026-10-15 | - |
| FORTRESS (Benign) | 90% | Thinking Machines, Thinking Machines Inkling model cardchecked 2026-10-15 | - |
| StrongREJECT | 98.5% | Thinking Machines, Thinking Machines Inkling model cardchecked 2026-10-15 | - |
| B3 AI Security Benchmark (attack resistance) | 88.1% | Mistral AI, Mistral Large 4 (preview)checked 2026-10-15 | - |
| Cyber refusal rate (average of JailbreakBench, StrongREJECT, AgentHarm) | 90% | Mistral AI, Mistral Large 4 (preview)checked 2026-10-15 | - |
| KORA Benchmark (0 to 2) | 1.4 | Mistral AI, Mistral Large 4 (preview)checked 2026-10-15 | - |
Preference and Composite Rankings
| Test | Value | Printed by | Setup as printed |
|---|---|---|---|
| Surge AI human eval on code (1 to 5) | 3.4 | Mistral AI, Mistral Large 4 (preview)checked 2026-10-15 | - |
Price
API list prices in US dollars per one million tokens, as printed on the developer's pricing page on the day checked. Cost per task is measured by Vals AI.
| Price | Value | Printed by | Setup as printed |
|---|---|---|---|
| Input | $1.4 | Z.ai, Z.AI API pricingchecked 2026-10-15 | - |
| Cached input | $0.26 | Z.ai, Z.AI API pricingchecked 2026-10-15 | - |
| Output | $4.4 | Z.ai, Z.AI API pricingchecked 2026-10-15 | - |
| Cost per taskVals AI, USD, lower is better | $8.49 | Vals AI, Vibe Code Bench v1.1 leaderboardchecked 2026-10-27 | Harness: OpenHands |
More from Z.ai
Every number here is read from the same catalog file as the benchmark table, data version 2026103001. Where one page printed the model under several setups, each value is listed. The catalog is public in armanobosyan/sugra-llm-benchmarks; a wrong number can be reported from the table, with Report an error under the value, or as an issue in that repository.