Gemini 4 Hydrogen Benchmarks: How It Compares Against Sonnet 5.5 and GPT-6.1 Terra
Google has officially announced Gemini 4 Hydrogen, the latest model in the Gemini 4 family. Gemini 4 Hydrogen is built on Gemini 4 Carbon, but it's faster, and it's a big upgrade over Gemini 3.8 Flash. Here is how it compares with other fast and efficient models, Claude Sonnet 5.5 and GPT-6.1 Terra, on the benchmarks Google shared.
All benchmarks
| Benchmark | Gemini 4 Hydrogen | Claude Sonnet 5.5 | GPT-6.1 Terra |
|---|---|---|---|
| Agentic coding: SWE-sweep | 13.5% | 3.1% | 5.8% |
| Terminal-Bench 4.1, cheating-adjusted | 57.0% | 63.9% | - |
| Knowledge work: Agents' Last Exam v1.1 | 71.6% | 60.4% | 69.8% |
| AutomationBench-AA | 76.1% | 71.8% | 60.2% |
| Vending-Bench 2, bank balance | $14,549 | $11,165 | $9,107 |
| Computer use: OSWorld 2.1, offline | 77.6% | 83.9% | 74.2% |
| Multimodal: MKA, Climb WSR | 53.4% | 52.0% | 43.8% |
| Chartography | 66.6% | 61.6% | 63.2% |
| Science: Terminal-Bench Science 0.1 | 51.4% | 59.9% | 58.3% |
| Reasoning: HLE-Diamond | 58.8% | 53.3% | 47.3% |
A dash means no score has been reported for that model.
Winner by benchmark
| Benchmark | Winner |
|---|---|
| SWE-sweep | Gemini 4 Hydrogen |
| Terminal-Bench 4.1 | Claude Sonnet 5.5 |
| Agents' Last Exam | Gemini 4 Hydrogen |
| AutomationBench | Gemini 4 Hydrogen |
| Vending-Bench 2 | Gemini 4 Hydrogen |
| OSWorld 2.1 | Claude Sonnet 5.5 |
| MKA | Gemini 4 Hydrogen |
| Chartography | Gemini 4 Hydrogen |
| Terminal-Bench Science | Claude Sonnet 5.5 |
| HLE-Diamond | Gemini 4 Hydrogen |
Gemini 4 Hydrogen vs Gemini 3.8 Flash
| Measure | Gemini 4 Hydrogen | Gemini 3.8 Flash |
|---|---|---|
| Terminal-Bench 4.1 | 57.0% | 18.2% |
| Task speed at the default setting | 2.5x faster | Baseline |
| Fabricated claims | More than 5x fewer | Baseline |
| Data-heavy enterprise workflows | 2x to 2.8x faster | Baseline |
| Tokens in those workflows | 12% to 23% fewer | Baseline |
The last two rows come from Google's early partner testing.
Artificial Analysis Intelligence Index
| Model | Score |
|---|---|
| Gemini 4 Hydrogen, xhigh | 54 |
| Gemini 4 Hydrogen, high (default) | 51 |
| Gemini 4 Argon | 50 |
Leaderboard positions
| Benchmark | Score | Position |
|---|---|---|
| Vals Tax Agent Bench | 52.3% | #1 |
| Vals Legal Research Bench | 55.8% | Tied #1 |
| Agent Arena | - | #5 |
API pricing
| Model | Input, per 1M tokens | Output, per 1M tokens |
|---|---|---|
| Gemini 4 Hydrogen | $1.50 | $7.50 |
| Claude Sonnet 5.5 | $2 | $10 |
| GPT-6.1 Terra | $1 | $5 |
Gemini 4 Hydrogen costs the same across its whole 1M-token context window.
Specifications
| Specification | Gemini 4 Hydrogen |
|---|---|
| Model ID | gemini-4-hydrogen |
| Base model | Adapted from Gemini 4 Carbon |
| Context window | 1M tokens, input and output combined |
| Maximum output | 64K tokens, up to 1M with background long decode |
| Knowledge cutoff | March 2026 |
| Text input | Yes |
| Image input | Yes |
| Audio input | Yes |
| Video input | Yes |
| Output | Text |
| Thinking levels | Low, Medium, High, xhigh |
| Default thinking | High |
| Release | November 12, 2026 |
Availability
| Platform | Gemini 4 Hydrogen |
|---|---|
| Gemini API | Free and paid tiers |
| Google AI Studio | Yes |
| Gemini app | Everyone, in the model picker |
| Google Search | All users, globally |
| Google Antigravity | Google AI Pro and Ultra |
| Gemini in Google Sheets | Google AI Pro and Ultra |
| Android Studio | Yes |
| Gemini for business | Yes |