Gemini 4 Argon vs GPT-6.1 Sol: Is OpenAI Falling Behind?
Google has made Gemini 4 Argon broadly available, and it edges past GPT-6.1 Sol on most benchmarks, including the Artificial Analysis Intelligence Index, 53 to 52. GPT-6.1 Sol still wins on cost: it runs the Index for less than half the price and keeps $2 and $10 pricing after Argon's introductory price ends on December 1. Here is how the two compare.
Benchmarks
| Benchmark | Gemini 4 Argon | GPT-6.1 Sol | Difference |
|---|---|---|---|
| DeepSWE v1.1 | 77.9% | 75.2% | Argon +2.7 points |
| AutomationBench | 51.3% | 36.1% | Argon +15.2 points |
| OSWorld 2.0 (offline, partial credit) | 69.2% | 71.4% | Sol +2.2 points |
| Terminal-Bench Science 0.1 | 57.6% | 57.0% | Argon +0.6 points |
Artificial Analysis benchmarks
| Benchmark | Gemini 4 Argon (High) | GPT-6.1 Sol (Max) | Difference |
|---|---|---|---|
| Intelligence Index | 53 | 52 | Argon +1 |
| AA-Briefcase v1.1 | 1488 Elo | 1561 Elo | Sol +73 Elo |
| GDPval-AA v2.1 | 1624 Elo | 1592 Elo | Argon +32 Elo |
| AutomationBench-AA | 78% | 65% | Argon +13 points |
| Terminal-Bench 4.0 | 57% | 56% | Argon +1 point |
| SciCode | 62% | 54% | Argon +8 points |
| Humanity's Last Exam | 57% | 53% | Argon +4 points |
| GDP.pdf | 22% | 31% | Sol +9 points |
| CritPt | 27% | 32% | Sol +5 points |
| AA-Omniscience | 42 | 42 | Tie |
| AA-LCR v1.1 | 80% | 83% | Sol +3 points |
Cost and efficiency
| Metric | Gemini 4 Argon (High) | GPT-6.1 Sol (Max) | Difference |
|---|---|---|---|
| Intelligence Index | 53 | 52 | Argon +1 |
| Cost per Intelligence Index task | $1.99 | $0.72 | Sol about 64% cheaper |
| Cost to run the Intelligence Index | $2,407 | $1,082 | Sol about 55% cheaper |
| Output tokens per task | 62K | 38K | Sol uses about 39% fewer |
| Reasoning tokens per task | 36K | 25K | Sol uses about 31% fewer |
| Output tokens across the Index | 113M | 67M | Sol uses about 41% fewer |
| Input, per 1M tokens | $2 | $2 | Same for now |
| Cached input, per 1M tokens | $0.10 | $0.10 | Same for now |
| Output, per 1M tokens | $10 | $10 | Same for now |
Pricing and specs
| Specification | Gemini 4 Argon | GPT-6.1 Sol | Difference |
|---|---|---|---|
| Input now, per 1M tokens | $2Introductory | $2 | Same |
| Cached input now, per 1M tokens | $0.10 | $0.10 | Same |
| Output now, per 1M tokens | $10Introductory | $10 | Same |
| Input from December 1, 2026 | $4 | $2 | Sol 50% cheaper |
| Output from December 1, 2026 | $20 | $10 | Sol 50% cheaper |
| Context window | About 1M tokens | 1.05M tokens | Nearly identical |
| Max output | Up to 1M tokens | 128K tokens | Argon much larger |
| Thinking levels | Low, Medium, High | Low, Medium, High, Xhigh, Max | Sol more granular |
| Text input | Yes | Yes | Same |
| Image input | Yes | Yes | Same |
| Audio and video input | Yes | No | Argon broader |
| API | Gemini API | OpenAI API | Both available |