Up to four models
Gemini 3.7 Flash vs GLM-5.3
On the index (GPT-4 = 100), Gemini 3.7 Flash scores 170 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.
170154–186
WeightsClosed
ReleasedAug 2026
Price in / out$0.75 / $3.75
Context1.05M
Core tests taken6 of 9
168152–184
WeightsOpen
ReleasedAug 2026
Price in / out$0.070 / $1.4
Context1.05M
Core tests taken5 of 9
Gemini 3.7 Flash and GLM-5.3 are too close to call (2 points apart, within the uncertainty), ahead on 4 of 5 shared core tests, at 3.7× the price.
Test by testthird-party results
Every result as a table
| Test | Gemini 3.7 Flash | GLM-5.3 |
|---|---|---|
| Reasoning & knowledge | ||
| GPQA Diamond (Epoch AI) | 94.8% | 90.9% |
| GPQA Diamond (Vals AI) | 93.9% | 88.1% |
| MMLU Pro | 90.1% | 86.8% |
| SimpleQA Verified | 69.2% | 41.0% |
| Maths | ||
| FrontierMath (tiers 1-3) | 71.6% | 68.8% |
| AIME-style maths (OTIS mock) | 97.2% | 91.1% |
| ProofBench v1.1 | 58.0% | 49.0% |
| Chess puzzles | 66.0% | 37.0% |
| FrontierMath (tier 4) | 36.6% | 29.3% |
| Coding | ||
| Terminal-Bench 4.0 | 12.1% | 38.9% |
| Code Migration | 34.8% | 44.2% |
| CyberBench v1.1 | 43.8% | 72.1% |
| IOI | 67.8% | 68.4% |
| LiveCodeBench | 88.6% | 80.5% |
| ProgramBench | 0.0% | 1.5% |
| SWE-bench | 80.8% | 95.4% |
| Vibe Code Bench v1.1 | 70.4% | 78.1% |
| DeepSWE | 65.5% | 69.0% |
| FrontierSWE | 20.3% | 30.2% |
| Agents | ||
| APEX-Agents | 67.8% | 56.6% |
| SkillsBench | 65.9% | 47.5% |
| SRE Bench | 4.6% | 9.2% |
| Terminal-Bench-Science | 7.1% | 5.7% |
| DTBench | 96.8% | 87.7% |
| LMCA | 50.4% | 55.5% |
| Mystery game puzzles | 37.0% | 33.0% |
| Professional work | ||
| Finance Agent (v2) | 59.0% | 55.8% |
| EMB | 71.3% | 56.3% |
| LegalBench | 87.3% | 84.8% |
| Legal Research Bench | 34.6% | 49.0% |
| MedCode | 53.4% | 42.9% |
| MedScribe | 83.9% | 88.8% |
| Tax Agent Bench | 22.8% | 43.3% |
| TaxEval v2 | 74.7% | 72.4% |
Tests at least two of these models have taken. Bold rows are the core tests.