GLM-5.3-Flash scores 40.4 on the Respan Index (#47), ahead of GLM-5.2 at 40.1 (#48). GLM-5.3-Flash leads on 16 of the 27 benchmarks both report. GLM-5.3-Flash is 9.1x cheaper per token at list price.
* Estimated: no published scores in that category. How the Respan Index works
| 59.2% |
| 55.8% |
| GPQA Diamond | 91.9% | 90.2% |
| LiveBench Agentic Coding (LiveBench) | 51.8% | 56.8% |
| LiveBench Coding (LiveBench) | 79.7% | 79% |
| LiveBench Data Analysis (LiveBench) | 73.7% | 76.4% |
| LiveBench Instruction Following (LiveBench) | 62.3% | 52.8% |
| LiveBench Language (LiveBench) | 76.2% | 77.3% |
| LiveBench Mathematics (LiveBench) | 89.8% | 81.2% |
| LiveBench Reasoning (LiveBench) | 78.6% | 77.6% |
| Mystery Game Puzzles (Epoch AI) | 19% | 8% |
| OTIS Mock AIME 2024-2025 (Epoch AI) | 86.4% | 93.9% |
| LMArena Agent (LMArena) | 0.0348 | -0.0044 |
| LMArena Elo | 1476 | 1472 |
| LMArena WebDev (LMArena) | 1605 | 1616 |
| Agents' Last Exam (ALE-CLI) | 23.8% | 26.3% |
| AutomationBench v1.0.6 | 26.2% | 48.8% |
| DeepSWE v1.1 | 46.2% | 63.4% |
| GDPval-AA v2 | 1508 | 1773 |
| HLE (with tools) | 54.7% | 55.3% |
| NL2Repo | 48.9% | 56.3% |
| Terminal-Bench 2.1 | 81% | 84.3% |
| Toolathlon Verified | 59.9% | 78.4% |
| Toolathlon-Verified (HKUST) | 59.9% | 78.4% |
Higher is better. Each score links to where it was published.