Up to four models
GLM-5.3 vs GPT-5.6 Luna
On the index (GPT-4 = 100), GLM-5.3 scores 172 and GPT-5.6 Luna 170. Side by side on the same third-party tests, with price and context. The link in your address bar shares this exact comparison.
GLM-5.3172155–188
WeightsOpen
ReleasedAug 2026
Price in / out$0.077 / $1.4
Context1.05M
Core tests taken5 of 9
GPT-5.6 Luna
170154–187
WeightsClosed
ReleasedJul 2026
Price in / out$0.20 / $1.2
Context1.05M
Core tests taken6 of 9
GLM-5.3 and GPT-5.6 Luna are too close to call (1 point apart, within the uncertainty), ahead on 3 of 5 shared core tests, at a similar price.
Test by testthird-party results
GLM-5.3GPT-5.6 LunaEvery result as a table
| Test | GLM-5.3 | GPT-5.6 Luna |
|---|---|---|
| Reasoning & knowledge | ||
| GPQA Diamond (Epoch AI) | 90.9% | 91.6% |
| LiveBench Data Analysis | 70.2% | 78.0% |
| LiveBench IF | 69.3% | 60.1% |
| LiveBench Language | 79.9% | 72.6% |
| LiveBench Reasoning | 85.8% | 85.6% |
| GPQA Diamond (Vals AI) | 88.1% | 91.7% |
| MMLU Pro | 86.8% | 86.0% |
| SimpleQA Verified | 41.0% | 41.0% |
| Maths | ||
| FrontierMath (tiers 1-3) | 68.8% | 82.1% |
| AIME-style maths (OTIS mock) | 91.1% | 98.3% |
| LiveBench Mathematics | 87.9% | 87.2% |
| ProofBench v1.1 | 49.0% | 60.0% |
| Chess puzzles | 37.0% | 43.0% |
| FrontierMath (tier 4) | 29.3% | 61.0% |
| Coding | ||
| Terminal-Bench 4.0 | 38.9% | 11.6% |
| LiveBench Agentic Coding | 60.9% | 48.4% |
| LiveBench Coding | 79.0% | 82.9% |
| Code Migration | 44.2% | 44.5% |
| CyberBench v1.1 | 72.1% | 73.6% |
| IOI | 68.4% | 61.8% |
| ProgramBench | 1.5% | 0.0% |
| SWE-bench | 95.4% | 93.0% |
| Vibe Code Bench 1-100 | 20.0% | 22.6% |
| Vibe Code Bench v1.1 | 78.1% | 77.1% |
| DeepSWE | 69.0% | 67.2% |
| WeirdML | 75.4% | 60.9% |
| Agents | ||
| APEX-Agents | 56.6% | 43.0% |
| SkillsBench | 47.5% | 60.5% |
| SRE Bench | 9.2% | 1.5% |
| Terminal-Bench-Science | 5.7% | 0.0% |
| DTBench | 87.7% | 89.1% |
| LMCA | 55.5% | 48.5% |
| Mystery game puzzles | 33.0% | 21.0% |
| Professional work | ||
| Finance Agent (v2) | 55.8% | 55.0% |
| EMB | 56.3% | 67.1% |
| LegalBench | 84.8% | 84.0% |
| Legal Research Bench | 49.0% | 36.5% |
| MedCode | 42.9% | 42.4% |
| MedScribe | 88.8% | 84.4% |
| Public Benefits Bench v1.1 | 68.5% | 61.2% |
| Tax Agent Bench | 39.7% | 24.8% |
| TaxEval v2 | 72.4% | 76.2% |
| Novel problems | ||
| MysteryMechanism | 23.0% | 14.4% |
Tests at least two of these models have taken. Bold rows are the core tests.
