Up to four models
Gemini 4 Argon vs Gemini 3.8 Flash
On the index (GPT-4 = 100), Gemini 4 Argon scores 195 and Gemini 3.8 Flash 175. Side by side on the same third-party tests, with price and context. The link in your address bar shares this exact comparison.
Gemini 4 Argon195179–212
WeightsClosed
ReleasedSep 2026
Price in / out$2.0 / $10
Context1.05M
Core tests taken6 of 9
Gemini 3.8 Flash175159–191
WeightsClosed
ReleasedSep 2026
Price in / out$0.82 / $4.13
Context1.05M
Core tests taken7 of 9
Gemini 4 Argon leads Gemini 3.8 Flash by 20 points, ahead on 6 of 6 shared core tests, at 2.4× the price.
Test by testthird-party results
Gemini 4 ArgonGemini 3.8 FlashEvery result as a table
| Test | Gemini 4 Argon | Gemini 3.8 Flash |
|---|---|---|
| Reasoning & knowledge | ||
| GPQA Diamond (Epoch AI) | 96.4% | 95.4% |
| SimpleQA Verified | 76.0% | 69.7% |
| Maths | ||
| FrontierMath (tiers 1-3) | 86.0% | 68.4% |
| AIME-style maths (OTIS mock) | 100.0% | 98.9% |
| ProofBench v1.1 | 99.0% | 48.0% |
| Chess puzzles | 59.0% | 58.0% |
| FrontierMath (tier 4) | 78.0% | 21.9% |
| Coding | ||
| Terminal-Bench 4.0 | 57.6% | 19.7% |
| Code Migration | 68.2% | 36.5% |
| CyberBench v1.1 | 77.9% | 43.8% |
| IOI | 100.0% | 56.9% |
| ProgramBench | 2.5% | 1.0% |
| Vibe Code Bench 1-100 | 21.2% | 18.8% |
| Vibe Code Bench v1.1 | 91.9% | 78.6% |
| FrontierSWE | 55.0% | 19.6% |
| Agents | ||
| APEX-Agents | 82.4% | 64.3% |
| BioMysteryBench | 76.3% | 62.2% |
| SRE Bench | 44.3% | 11.1% |
| Terminal-Bench-Science | 44.3% | 8.6% |
| DTBench | 97.6% | 95.7% |
| Furniture assembly | 72.5% | 31.7% |
| LMCA | 63.7% | 52.9% |
| Mystery game puzzles | 64.0% | 47.0% |
| Professional work | ||
| Finance Agent (v2) | 65.4% | 61.4% |
| EMB | 75.2% | 72.2% |
| HLAB-Vals-1.1 | 27.1% | 12.1% |
| LegalBench | 88.3% | 87.0% |
| Legal Research Bench | 54.8% | 38.9% |
| MedCode | 58.8% | 48.1% |
| MedScribe | 87.4% | 84.5% |
| Public Benefits Bench v1.1 | 69.8% | 65.3% |
| SAGE | 53.6% | 35.1% |
| Tax Agent Bench | 44.6% | 32.1% |
| Novel problems | ||
| ARC-AGI-2 | 91.3% | 89.2% |
| ARC-AGI-1 | 98.5% | 98.5% |
| ARC-AGI-3 | 49.8% | 10.4% |
| MysteryMechanism | 45.5% | 36.5% |
Tests at least two of these models have taken. Bold rows are the core tests.
