Gemini 4 Argon vs Gemini 3.8 Flash
Finally, a pro model from Google Gemini. Google has officially announced Gemini 4 Argon, and here is how it compares with Gemini 3.8 Flash, which was announced recently.
Benchmarks
| Benchmark | Gemini 4 Argon | Gemini 3.8 Flash | Difference |
|---|---|---|---|
| DeepSWE v1.1 | 77.9% | 73.7% | Argon +4.2 points |
| Vals Finance Agent v2 | 65.4% | 61.4% | Argon +4.0 points |
| Harvey Legal Agent | 19.6% | 10.0% | Argon +9.6 points |
| Terminal-Bench 4.0 | 57.4% | 19.1% | Argon +38.3 points |
| LABBench 2 | 88.8% | 86.2% | Argon +2.6 points |
| OSWorld 2.0 (partial credit) | 69.2% | 59.0% | Argon +10.2 points |
| LVBench | 91.7% | 87.8%Agentic | Argon +3.9 points |
Artificial Analysis benchmarks
| Benchmark | Gemini 4 Argon (High) | Gemini 3.8 Flash (High) | Difference |
|---|---|---|---|
| Intelligence Index v4.3.2 | 53 | 41 | Argon +12 |
| AA-Briefcase v1.1 | 1494 Elo | 1202 Elo | Argon +292 Elo |
| GDPval-AA v2.1 | 1611 Elo | 1412 Elo | Argon +199 Elo |
| AutomationBench-AA | 78% | 60% | Argon +18 points |
| Terminal-Bench 4.0 | 57% | 20% | Argon +37 points |
| SciCode | 62% | 57% | Argon +5 points |
| Humanity's Last Exam | 57% | 48% | Argon +9 points |
| GDP.pdf | 22% | 21% | Argon +1 point |
| CritPt | 27% | 18% | Argon +9 points |
| AA-Omniscience | 42 | 30 | Argon +12 |
| AA-LCR v1.1 | 80% | 81% | 3.8 Flash +1 point |
Cost and efficiency
| Metric | Gemini 4 Argon (High) | Gemini 3.8 Flash (High) | Difference |
|---|---|---|---|
| Intelligence Index | 53 | 41 | Argon +12 |
| Cost per Intelligence Index task | $1.99 | $1.24 | 3.8 Flash about 38% cheaper |
| Output tokens per task | 62K | 71K | Argon uses about 13% fewer |
| Reasoning tokens per task | 36K | 43K | Argon uses about 16% fewer |
| Output tokens across the Index | 113M | 172M | Argon uses about 34% fewer |
| Input, per 1M tokens | $2Introductory | $0.75Introductory | 3.8 Flash 62.5% cheaper |
| Output, per 1M tokens | $10Introductory | $3.75Introductory | 3.8 Flash 62.5% cheaper |
| Cached input, per 1M tokens | $0.10Introductory | $0.075Introductory | 3.8 Flash 25% cheaper |
Pricing and specs
| Specification | Gemini 4 Argon | Gemini 3.8 Flash | Difference |
|---|---|---|---|
| Input now, per 1M tokens | $2Introductory | $0.75Introductory | 3.8 Flash about 2.7x cheaper |
| Cached input now, per 1M tokens | $0.10Introductory | $0.075Introductory | 3.8 Flash cheaper |
| Output now, per 1M tokens | $10Introductory | $3.75Introductory | 3.8 Flash about 2.7x cheaper |
| Input later, per 1M tokens | $4Standard | $1.50Standard | 3.8 Flash about 2.7x cheaper |
| Output later, per 1M tokens | $20Standard | $7.50Standard | 3.8 Flash about 2.7x cheaper |
| Context window | 1M tokens | 1M tokens | Same |
| Max output | 1M tokens | 64K tokens | Argon about 15.6x larger |
| Thinking | Yes | Yes, Low, Medium and High | Both |
| Public API status | Limited rollout | Generally available | 3.8 Flash is wider |
| Released | September 30, 2026 | September 2, 2026 | Argon is 28 days newer |