Gemini 4 Hydrogen
Everyday execution
- Announced
- Nov 12, 2026
- Per 1M tokens, in / out
- $1.50 / $7.50
- Thinking levels
- low → xhigh
Periodic family
All four read a 1,048,576-token context.
Gemini 4 Hydrogen
Everyday execution
Gemini 4 Carbon
The hardest problems
Gemini 4 Neon
Depth with efficiency
Gemini 4 Argon
Long-horizon reasoning
C · 6 · Mohs 10
Google’s strongest model, built for the hardest problems.
$2 in / $10 out per 1M tokens, introductory until Nov 30, 2026, then $4 / $20
Valence 4 · sp³ 109.5°
/swarm in Google Antigravity, in preview
Google’s /swarm demo: a reconstruction of Platform 37
Diamond · the hardest natural material
Project Fortify: 100% recall, 95% precision. Wiz Scan for Good: 0-days found in under two hours.
C–C bond 154 pm · the element of life
CASP17 blind test set, seven low-homolog families
Type IIa · flawless clarity
Trajectory Labs: no new universal jailbreaks across 132 hours and 13,000+ requests.
CH₄ · tetrahedral 109.5°
Orchestrator in Google Antigravity switches between the two inside one session.
Terminal-Bench 4.0: near-Carbon task success with 24% less quota, 19% faster on average.
H · 1 · 1.008 u
The daily driver for agentic work.
The lightest element
than Gemini 3.8 Flash, at default thinking
Agentic tool use
Batches file reads and searches, and adapts when a call fails.
Embodied reasoning
Drives a 6-axis arm from a plain-language goal and a camera view.
The most abundant element
Ne · 10 · 640.2 nm
Depth and efficiency, in balance.
Steerability
Move your pointer across the tube to steer it.
Noble gas · full shell 2s² 2p⁶
Lowest fabrication rate recorded so far in Google’s internal Antigravity testing.
Long-running agents
Runs parallel Hydrogen subagents, rereads fewer files, compacts its context.
Ar · 18 · 39.948 u
Built to sustain deep reasoning across long-horizon work.
Argon · welding shield gas
Leading on Gray Swan’s indirect prompt injection benchmark at launch.
Ar⁺ laser · 488 nm
Long decode
Up from 64K, sixteen times longer.
Spectrograph
The Gemini 4 Carbon evaluation table: 17 benchmarks, 5 models. Each score is an emission line.
Scroll, or use ↑ ↓
| Benchmark | Gemini 4 Carbon | Gemini 4 Argon | GPT-6 Astra | Claude Opus 5.5 | Claude Fable 5.1 |
|---|---|---|---|---|---|
| General capabilities | |||||
| AA Index v5.0 | 59 | 50 | 57 | 58 | 49 |
| Vals Index v2.2 | 70.4% | 69.2% | 65.0% | 70.2% | — |
| Agentic coding | |||||
| SWE-sweep | 9.6% | 4.3% | 9.5% | 3.6% | 2.9% |
| FrontierSWE 2 | 61.2% | 55.0% | 65.5% | 62.3% | 56.3% |
| Terminal-Bench 4.0AA | 65.2% | 57.1% | 59.1% | 59.6% | 52.0% |
| Computer use | |||||
| Agents' Last Exam v1.1Score | 77.4% | 69.0% | 73.1% | 71.6% | — |
| OSWorld-2.1Offline subset | 81.5% | 74.7% | 80.9% | 87.2% | 76.8% |
| Knowledge work | |||||
| AutomationBenchZapier | 53.9% | 51.3% | 41.4% | 42.5% | 31.4% |
| Vending-Bench 2Bank balance ($) | $16,484 | $13,718 | $15,515 | $9,235 | $5,422 |
| Multimodal understanding | |||||
| Chartography | 72.5% | 71.6% | 71.0% | 66.3% | 46.2% |
| Blueprint-Bench 2Similarity score | 58.6% | 54.4% | 49.7% | 51.2% | 41.9% |
| Long context | |||||
| GraphWalks256k to 1M, BFS (F1) | 96.2% | 84.2% | 71.8% | 66.8% | 65.0% |
| Science and math | |||||
| RiemannBench | 80.0% | 76.0% | 72.0% | 69.6% | 65.6% |
| Terminal-Bench-Science 0.1AA | 64.8% | 52.9% | 63.3% | 59.0% | 43.3% |
| Cybersecurity | |||||
| CWE-bench | 71.0% | 68.0% | 68.0% | 67.0% | 58.0% |
| ExploitGym | 54.8% | 42.0% | 42.4% | 34.5% | — |
| SEC-bench Pro | 88.2% | 83.4% | 85.4% | — | — |
Methodology: deepmind.google/models/evals-methodology/gemini-4-carbon
Hydrogen · Carbon · Neon · Argon
Gemini API · Google AI Studio · Google Antigravity · Gemini app