Up to four models
Grok 4.7 vs Gemini 3.1 Pro
On the index (GPT-4 = 100), Grok 4.7 scores 168 and Gemini 3.1 Pro 166. Side by side on the same third-party tests, with price and context. The link in your address bar shares this exact comparison.
Grok 4.7
168151–184
WeightsClosed
ReleasedSep 2026
Price in / out$2.0 / $6.0
Context500K
Core tests taken6 of 9
Gemini 3.1 Pro166150–183
WeightsClosed
ReleasedFeb 2026
Price in / out$2.0 / $12
Context1.05M
Core tests taken9 of 9
Grok 4.7 and Gemini 3.1 Pro are too close to call (1 point apart, within the uncertainty), ahead on 3 of 6 shared core tests, and costs 1.5× less.
Test by testthird-party results
Grok 4.7Gemini 3.1 Pro
Gemini 3.1 ProEvery result as a table
| Test | Grok 4.7 | Gemini 3.1 Pro |
|---|---|---|
| Reasoning & knowledge | ||
| GPQA Diamond (Epoch AI) | 92.7% | 94.4% |
| SimpleQA Verified | 56.0% | 73.5% |
| Maths | ||
| FrontierMath (tiers 1-3) | 53.0% | 59.7% |
| AIME-style maths (OTIS mock) | 98.1% | 95.6% |
| ProofBench v1.1 | 26.0% | 26.0% |
| Chess puzzles | 41.0% | 51.0% |
| FrontierMath (tier 4) | 17.1% | 26.8% |
| Coding | ||
| Terminal-Bench 4.0 | 28.8% | 2.5% |
| Code Migration | 44.8% | 17.3% |
| IOI | 57.7% | 51.8% |
| ProgramBench | 0.5% | 0.0% |
| Vibe Code Bench v1.1 | 86.2% | 32.0% |
| Agents | ||
| APEX-Agents | 54.6% | 35.1% |
| BioMysteryBench | 69.3% | 60.0% |
| SRE Bench | 3.8% | 0.0% |
| Terminal-Bench-Science | 10.0% | 1.4% |
| DTBench | 96.0% | 97.1% |
| Furniture assembly | 20.8% | 26.7% |
| LMCA | 49.4% | 53.8% |
| Mystery game puzzles | 29.0% | 34.0% |
| Professional work | ||
| Finance Agent (v2) | 52.3% | 43.0% |
| EMB | 67.0% | 52.6% |
| LegalBench | 84.4% | 87.4% |
| Legal Research Bench | 47.1% | 20.7% |
| MedCode | 49.5% | 59.1% |
| MedScribe | 89.4% | 76.1% |
| Public Benefits Bench v1.1 | 65.6% | 53.8% |
| SAGE | 40.8% | 48.7% |
| Tax Agent Bench | 34.2% | 9.3% |
| Novel problems | ||
| ARC-AGI-2 | 61.4% | 77.1% |
| ARC-AGI-1 | 90.2% | 98.0% |
| ARC-AGI-3 | 1.8% | 0.4% |
| MysteryMechanism | 25.2% | 20.7% |
Task length · METR46 sec
46 secTests at least two of these models have taken. Bold rows are the core tests.
