Up to four models
GLM-5.3 vs Gemini 3.6 Flash
On the index (GPT-4 = 100), GLM-5.3 scores 168 and Gemini 3.6 Flash 161. Side by side on the same third-party tests, with price and context. The link in your address bar shares this exact comparison.
168152–184
WeightsOpen
ReleasedAug 2026
Price in / out$0.070 / $1.4
Context1.05M
Core tests taken5 of 9
161145–176
WeightsClosed
ReleasedJul 2026
Price in / out$0.75 / $3.75
Context1.05M
Core tests taken5 of 9
GLM-5.3 leads Gemini 3.6 Flash by 8 points, ahead on 2 of 4 shared core tests, and costs 3.7× less.
Test by testthird-party results
Every result as a table
| Test | GLM-5.3 | Gemini 3.6 Flash |
|---|---|---|
| Reasoning & knowledge | ||
| GPQA Diamond (Epoch AI) | 90.9% | 94.1% |
| GPQA Diamond (Vals AI) | 88.1% | 93.4% |
| MMLU Pro | 86.8% | 89.3% |
| SimpleQA Verified | 41.0% | 66.2% |
| Maths | ||
| FrontierMath (tiers 1-3) | 68.8% | 59.0% |
| AIME-style maths (OTIS mock) | 91.1% | 94.2% |
| Chess puzzles | 37.0% | 42.0% |
| FrontierMath (tier 4) | 29.3% | 21.9% |
| Coding | ||
| Code Migration | 44.2% | 30.9% |
| CyberBench v1.1 | 72.1% | 45.4% |
| IOI | 68.4% | 35.1% |
| LiveCodeBench | 80.5% | 88.1% |
| ProgramBench | 1.5% | 0.0% |
| SWE-bench | 95.4% | 79.6% |
| Vibe Code Bench v1.1 | 78.1% | 64.0% |
| DeepSWE | 69.0% | 46.7% |
| WeirdML | 75.4% | 56.1% |
| Agents | ||
| APEX-Agents | 56.6% | 46.9% |
| Terminal-Bench-Science | 5.7% | 4.3% |
| DTBench | 87.7% | 95.5% |
| LMCA | 55.5% | 44.9% |
| Mystery game puzzles | 33.0% | 30.0% |
| Professional work | ||
| Finance Agent (v2) | 55.8% | 56.3% |
| EMB | 56.3% | 65.4% |
| LegalBench | 84.8% | 86.7% |
| Legal Research Bench | 49.0% | 25.0% |
| MedCode | 42.9% | 53.1% |
| MedScribe | 88.8% | 79.7% |
| Public Benefits Bench v1.1 | 68.5% | 56.6% |
| Tax Agent Bench | 43.3% | 18.0% |
| TaxEval v2 | 72.4% | 74.9% |
Tests at least two of these models have taken. Bold rows are the core tests.