GLM-5.2 vs GLM-5.3
GLM-5.2 Verdict
- GLM-5.3 (max) leads on quality: 65.8 vs 63.9.
- GLM-5.2 (max) is stronger in agents & tools, instruction following, reasoning.
- GLM-5.3 (max) is stronger in coding, composite, human preference, knowledge, long context, maths.
- They cost about the same ($2.15 per 1M blended).
| Metric | GLM-5.2 (max) | GLM-5.3 (max) |
|---|---|---|
| BenchLeader Index | 63.9 | 65.8 |
| Agents & tools score | 63.8 | 62.0 |
| Coding score | 63.8 | 64.4 |
| Composite score | 72.1 | 85.9 |
| Human preference score | 67.2 | 68.5 |
| Instruction following score | 71.6 | – |
| Knowledge score | 55.5 | 59.6 |
| Long context score | 65.6 | 66.3 |
| Maths score | 59.0 | 62.7 |
| Reasoning score | 72.9 | 71.6 |
| Blended price $/M | $2.15 | $2.15 |
| Output speed | 71 tok/s | 63 tok/s |
| Time to first answer | 31.6 s | 34.6 s |
| Context window | 1M | 1M |
| GPQA Diamond | 91.9% | 90.9% |
| FrontierMath Tiers 1–3 | 59.2% | 68.8% |
| FrontierMath Tier 4 | 29.3% | 29.3% |
| OTIS Mock AIME | 86.4% | 91.1% |
| SWE-bench Verified (Epoch) | 78.7% | – |
| SimpleQA Verified | 34.2% | 41.0% |
| Terminal-Bench | – | 41.8% |
| SciCode | 50.5% | 56.5% |
| WeirdML | 70.1% | 75.4% |
| ProofBench | 35.0% | 49.0% |
| LMArena Text | 1472 | 1483 |
| LMArena Hard Prompts | 1493 | 1507 |
| LMArena Coding | 1510 | 1524 |
| LMArena WebDev | 1592 | 1614 |
| LMArena Agent | 4.4 | 3.1 |
| AA Intelligence Index | 34.0 | 44.9 |
| IFBench | 73.3% | – |
| AA-LCR | 78.3% | 79.7% |
| AA-Omniscience | 4.4 | 14.3 |
| Terminal-Bench Hard | 50.8% | – |
| GPQA Diamond (AA) | 89.5% | 91.7% |
| Humanity's Last Exam (AA) | 41.1% | 42.3% |
| SciCode (AA) | 51.2% | 59.0% |
| τ²-Bench Telecom (AA) | 99.1% | – |
| LiveCodeBench | – | 80.5% |
| MMLU-Pro | – | 86.8% |
| IOI | – | 68.4% |
| LegalBench | – | 84.8% |
| TaxEval | – | 72.4% |
| Terminal-Bench 2.1 (Vals) | 67.8% | 71.5% |
| SWE-bench (Vals) | 82.8% | 95.4% |
| GPQA Diamond (Vals) | – | 88.1% |
| Vals Index | – | 57.0 |
| CritPt | 20.9% | 19.1% |
| GDPval (AA) | 45.3% | 56.7% |
| τ²-Bench Banking (AA) | 34.6% | 50.3% |
| ITBench SRE (AA) | 42.7% | – |
| APEX-Agents (AA) | 33.7% | – |
| Code Migration | 37.9% | 44.2% |
| Excel Modeling Benchmark | – | 56.3% |
| Finance Agent v2 | – | 55.8% |
| Harvey's Legal Agent Benchmark | 7.1% | 8.3% |
| Legal Research Bench | 31.3% | 49.0% |
| MedCode | – | 42.9% |
| MedScribe | – | 88.8% |
| MysteryMechanism | – | 23.0% |
| ProgramBench | 0.5% | 1.5% |
| Public Benefits Bench | – | 68.5% |
| SkillsBench | 45.1% | 47.5% |
| SREBench | 0.0% | – |
| Tax Agent Bench | – | 73.1% |
| Terminal-Bench 4.0 (Vals) | – | 25.3% |
| Terminal-Bench Science | – | 4.3% |
| Vibe Code Bench 1-100 | – | 20.0% |
| Vibe Code Bench v1.1 | 64.0% | 78.1% |
| FORTRESS | – | 28.2% |
| LMArena Maths | 1480 | 1504 |
| LMArena Creative Writing | 1451 | 1462 |
| LMArena Instruction Following | 1466 | 1480 |
| LMArena Multi-turn | 1469 | 1493 |
| LMArena Longer Queries | 1483 | 1492 |
| Chess Puzzles | 21.0% | 21.0% |
| EBR-bench | 9.5% | – |
| Mystery Game Puzzles | – | 33.0% |
| PostTrainBench | 31.7% | – |
| DeepSWE | 43.8% | 69.0% |
| LMCA | 45.8% | – |
| DTBench | 93.6% | – |
| CursorBench | 55.0% | – |
| ALE-Bench | 1010.2 | – |
| FrontierSWE | – | 30.2% |
Data as of 2026-09-19. Best configuration of each model; every score links to its source on the model pages.
GLM-5.2 vs GLM-5.3: questions
- Is GLM-5.2 better than GLM-5.3?
- GLM-5.3 (max) leads on quality: 65.8 vs 63.9. The BenchLeader Index combines every independent quality benchmark; GLM-5.3 (max) is ahead overall as of 2026-09-19, but check the category scores for your use.
- Is GLM-5.2 better than GLM-5.3 for coding?
- GLM-5.3 scores higher in coding (64 vs 64 on the category index, where 50 is average).
- Is GLM-5.2 better than GLM-5.3 for agentic tasks?
- GLM-5.2 scores higher in agentic tasks (64 vs 62 on the category index, where 50 is average).
- Which is cheaper, GLM-5.2 or GLM-5.3?
- GLM-5.2 is cheaper: $2.15 against $2.15 per million tokens, blended at three input tokens per output token.
- Which is faster, GLM-5.2 or GLM-5.3?
- GLM-5.2 streams faster: 71 against 63 output tokens per second.
- Which has the larger context window?
- Both accept 1M tokens of context.