Up to four models
GPT-6 Astra vs GLM-5.3
On the index (GPT-4 = 100), GPT-6 Astra scores 193 and GLM-5.3 168. Side by side on the same third-party tests, with price and context. The link in your address bar shares this exact comparison.
GPT-6 Astra
193178–209
WeightsClosed
ReleasedSep 2026
Price in / out$10 / $50
Context1.05M
Core tests taken7 of 9
168152–184
WeightsOpen
ReleasedAug 2026
Price in / out$0.070 / $1.4
Context1.05M
Core tests taken5 of 9
GPT-6 Astra leads GLM-5.3 by 25 points, ahead on 4 of 5 shared core tests, at 50× the price.
Test by testthird-party results
GPT-6 Astra
GLM-5.3
Every result as a table
| Test | GPT-6 Astra | GLM-5.3 |
|---|---|---|
| Reasoning & knowledge | ||
| GPQA Diamond (Epoch AI) | 95.8% | 90.9% |
| LiveBench Data Analysis | 83.0% | 70.2% |
| LiveBench IF | 75.6% | 69.3% |
| LiveBench Language | 89.4% | 79.9% |
| LiveBench Reasoning | 92.7% | 85.8% |
| SimpleBench | 83.6% | 66.2% |
| SimpleQA Verified | 75.6% | 41.0% |
| Maths | ||
| FrontierMath (tiers 1-3) | 93.7% | 68.8% |
| AIME-style maths (OTIS mock) | 100.0% | 91.1% |
| LiveBench Mathematics | 96.8% | 87.9% |
| ProofBench v1.1 | 99.0% | 49.0% |
| Chess puzzles | 74.0% | 37.0% |
| FrontierMath (tier 4) | 97.6% | 29.3% |
| Coding | ||
| Terminal-Bench 4.0 | 59.6% | 38.9% |
| LiveBench Agentic Coding | 57.3% | 60.9% |
| LiveBench Coding | 80.4% | 79.0% |
| Code Migration | 67.7% | 44.2% |
| CyberBench v1.1 | 41.1% | 72.1% |
| IOI | 100.0% | 68.4% |
| ProgramBench | 5.5% | 1.5% |
| Vibe Code Bench 1-100 | 27.6% | 20.0% |
| Vibe Code Bench v1.1 | 89.6% | 78.1% |
| DeepSWE | 74.1% | 69.0% |
| FrontierSWE | 65.5% | 30.2% |
| WeirdML | 93.3% | 75.4% |
| Agents | ||
| APEX-Agents | 64.7% | 56.6% |
| SRE Bench | 56.9% | 9.2% |
| Terminal-Bench-Science | 62.9% | 5.7% |
| DTBench | 97.3% | 87.7% |
| LMCA | 64.4% | 55.5% |
| Mystery game puzzles | 84.0% | 33.0% |
| Professional work | ||
| Finance Agent (v2) | 53.5% | 55.8% |
| EMB | 71.7% | 56.3% |
| HLAB-Vals-1.1 | 7.1% | 14.6% |
| Legal Research Bench | 39.4% | 49.0% |
| MedCode | 48.5% | 42.9% |
| MedScribe | 87.9% | 88.8% |
| Tax Agent Bench | 20.7% | 43.3% |
| Novel problems | ||
| MysteryMechanism | 53.1% | 23.0% |
Tests at least two of these models have taken. Bold rows are the core tests.