GLM-5.2 vs GLM-5.3
Respan Index by category
Each category is scored 0 to 100, and how far ahead the winner is. Leads under a point are close to even.
| Category | GLM-5.2 | GLM-5.3 |
|---|---|---|
| Coding | 27.9 GLM-5.2 | 44.4 GLM-5.3 |
| Agents | 55.9 GLM-5.2 | 64.6 GLM-5.3 |
| Reasoning | 42.0 GLM-5.2 | 48.5 GLM-5.3 |
| Knowledge & language | 34.6 GLM-5.2 | 42.4 GLM-5.3 |
| Human preference | 56.4 GLM-5.2 | 56.8 GLM-5.3 |
Benchmarks both report
GLM-5.3 leads on 33 of the 39 benchmarks both report. Higher is better. Each score links to where it was published.
| Benchmark | GLM-5.2 | GLM-5.3 |
|---|---|---|
| LMArena Elo | 1475 | 1482 |
| LMArena WebDev (LMArena) | 1608 | 1622 |
| LMArena Agent (LMArena) | -0.0041 | -0.0004 |
| Chess Puzzles (Epoch AI) | 22% | 37% |
| APEX-Agents (Mercor) | 45.2% | 56.6% |
| DeepSWE v1.1 (Datacurve) | 43.8% | 69% |
| FrontierMath Tier 4 v2 (Epoch AI) | 29.3% | 29.3% |
| FrontierMath Tiers 1-3 v2 (Epoch AI) | 59.2% | 68.8% |
| GPQA Diamond | 91.9% | 90.9% |
| LiveBench Agentic Coding (LiveBench) | 51.8% | 60.9% |
| LiveBench Coding (LiveBench) | 79.7% | 79% |
| LiveBench Data Analysis (LiveBench) | 73.7% | 70.2% |
| LiveBench Instruction Following (LiveBench) | 62.3% | 69.3% |
| LiveBench Language (LiveBench) | 76.2% | 79.9% |
| LiveBench Mathematics (LiveBench) | 89.8% | 87.9% |
| LiveBench Reasoning (LiveBench) | 78.6% | 85.8% |
| Mystery Game Puzzles (Epoch AI) | 19% | 33% |
| OTIS Mock AIME 2024-2025 (Epoch AI) | 86.4% | 91.1% |
| SimpleQA Verified | 34.2% | 41% |
| Vending-Bench 2 (Andon Labs) | 8313.78 | 8163.61 |
| WeirdML (Håvard Tveit Ihle) | 70.1% | 75.4% |
| Agents' Last Exam (ALE-CLI) | 23.8% | 28.5% |
| AutomationBench v1.0.6 | 26.2% | 48.2% |
| CyberGym | 77.2% | 84.5% |
| DeepSWE v1.1 | 46.2% | 66.9% |
| ExploitBench | 24.4% | 54.4% |
| ExploitGym 2h | 29 | 105 |
| ExploitGym 6h | 39 | 130 |
| FrontierSWE | 67.5% | 78.1% |
| GDPval-AA v2 | 1508 | 1769 |
| HLE (with tools) | 54.7% | 62.5% |
| NL2Repo | 48.9% | 58% |
| PostTrainBench | 31.7% | 39.8% |
| ProgramBench (Almost Solved) | 9.5% | 19% |
| SWE-Marathon v1.1 | 19.4% | 42.5% |
| Terminal-Bench 2.1 | 81% | 88.2% |
| Terminal-Bench 3.0 | 4.6% | 28.3% |
| Toolathlon Verified | 59.9% | 73% |
| SimpleBench (SimpleBench) | 58.8% | 66.2% |
FAQ
Is GLM-5.2 better than GLM-5.3?
GLM-5.3 scores 52.1 on the Respan Index (#39), ahead of GLM-5.2 at 43.4 (#62). GLM-5.3 leads on 33 of the 39 benchmarks both report.
Which is cheaper, GLM-5.2 or GLM-5.3?
GLM-5.2 lists at $0.19 input and $1.8 output per 1M tokens; GLM-5.3 at $0.07 and $1.4. GLM-5.3 is 1.5x cheaper per token at list price.
Which has the larger context window?
Both accept 1M tokens of context.