Gemini 4 Hydrogen vs GPT-6.1 Terra: Gemini 4 Hydrogen Has a Narrow Lead
Here is how Gemini 4 Hydrogen, Google's latest fast model, compares against GPT-6.1 Terra. Gemini 4 Hydrogen wins on overall performance, scoring 54 vs 51 on the Artificial Analysis Intelligence Index, and it does particularly well in automation and reasoning, with much faster output. GPT-6.1 Terra is still better at cost efficiency, and it remains the best cheaper model.
Google's benchmarks | |||
| Benchmark |  Gemini 4 Hydrogen |  GPT-6.1 Terra | Difference |
|---|---|---|---|
| SWE-sweep | 13.5% | 5.8% |  +7.7 pts |
| Terminal-Bench 4.1, cheating-adjusted | 57.0% | Not reported | |
| Agents' Last Exam v1.1 | 71.6% | 69.8% | +1.8 pts |
| AutomationBench-AA | 76.1% | 60.2% | +15.9 pts |
| Vending-Bench 2, bank balance | $14,549 | $9,107 | +$5,442 |
| OSWorld 2.1, offline | 77.6% | 74.2% | +3.4 pts |
| MKA, Climb WSR | 53.4% | 43.8% | +9.6 pts |
| Chartography | 66.6% | 63.2% | +3.4 pts |
| Terminal-Bench Science 0.1 | 51.4% | 58.3% |  +6.9 pts |
| HLE-Diamond | 58.8% | 47.3% | +11.5 pts |
Artificial Analysis benchmarks | |||
| Intelligence Index v5.0.1 | 54 at xhigh | 51 at max | +3 |
| AA-Briefcase v1.1 | 1567 Elo at xhigh | 1403 Elo at max | +164 Elo |
Cost and speed at top settings | |||
| Cost per Intelligence Index task | $3.16 | $0.96 | 70% cheaper |
| Output tokens across the Index | 304M | 99.4M | 67% fewer |
| Input tokens across the Index | 11.3B | 5.57B | 51% fewer |
| Output speed | 190 tokens/s | 68 tokens/s | 2.8x faster |
| Time to first answer token | 37.8 seconds | 170.4 seconds | 78% sooner |
| Time per Index task | 13.5 minutes | 14.6 minutes | 8% less |
Intelligence Index by effort | |||
| Max | Not offered | 51 | |
| xhigh | 54 | 45 | +9 |
| High | 51 | 42 | +9 |
| Medium | 41 | 38 | +3 |
| Low | 27 | 29 | +2 |
Cost per task by effort | |||
| Max | Not offered | $0.96 | |
| xhigh | $3.16 | $0.40 | 87% cheaper |
| High | $1.87 | $0.26 | 86% cheaper |
| Medium | $0.80 | $0.17 | 79% cheaper |
| Low | $0.29 | $0.07 | 76% cheaper |
AA-Briefcase by effort | |||
| Max | Not offered | 1403 Elo | |
| xhigh | 1567 Elo | 1333 Elo | +234 |
| High | 1497 Elo | 1264 Elo | +233 |
| Medium | 1296 Elo | 1190 Elo | +106 |
| Low | 947 Elo | 1031 Elo | +84 |
Pricing and specs | |||
| Input, per 1M tokens | $1.50 | $1 | 33% cheaper |
| Output, per 1M tokens | $7.50 | $5 | 33% cheaper |
| Cached input, per 1M tokens | $0.075 | $0.05 | 33% cheaper |
| Context window | 1M tokens (input + output) | 1.05M tokens | |
| Max output | 64K tokens, up to 1M with long decode | 128K tokens | 2x by default |
| Reasoning levels | Low to xhigh | None to max | |
| Default effort | High | Medium | |
| Inputs | Text, images, audio, video | Text, images | |
| Knowledge cutoff | March 2026 | June 2026 | 3 months newer |
| Released | November 12, 2026 | October 26, 2026 | 17 days newer |