1. VEGA
  2. Blog
  3. Announcing Gemini 4 Hydrogen
November 13, 2026Announcements

Announcing Gemini 4 Hydrogen

Announcing Gemini 4 Hydrogen

Google released Gemini 4 Hydrogen on November 12, 2026, two days after the broad rollout of Gemini 4 Carbon. Hydrogen takes over from the Flash line as Google's everyday model, and the Flash brand is gone with it — in the Gemini app the model now sits under the plain "Everyday help" label. Where Carbon chases deep, long-horizon reasoning, Hydrogen is tuned for fast execution inside agentic workflows.

What Hydrogen is

The model is distilled from the flagship Carbon and aimed at daily work: coding and debugging, interactive web development, research, and multimodal tasks. The specs:


  • Context: 1M tokens in, up to 1M tokens out

  • Thinking levels: low, medium, high (default), and a new xhigh tier

  • Modalities: text, image, audio, and video in; text out

  • Knowledge cutoff: March 2026, GA from day one

  • Gemini API pricing: $1.50 per million input tokens, $7.50 per million output

Google ships it everywhere at once: the Gemini app, AI Mode in Search globally (called the largest model upgrade in Search so far), AI Studio, the Gemini API, Gemini for business, and Antigravity for AI Pro and Ultra subscribers.

Speed, cost, reliability

At the default high setting Hydrogen finishes tasks 2.5× faster than Gemini 3.8 Flash. Partner tests on graph analytics, SQL, and data-heavy enterprise workflows measured 2–2.8× speedups while spending 12–23% fewer tokens.

Google also reports fabricated claims down more than 5× versus 3.8 Flash, and the model is built on the new Gemini 4 safety framework: resistance to indirect prompt injection improved more than 2.3× on the Gray Swan IPI benchmark, with added guardrails against agents overstepping permissions or executing malicious instructions.

Benchmarks

Hydrogen scores 51 on the Artificial Analysis Intelligence Index at high — edging out the larger and pricier Gemini 4 Argon at 50 — and reaches 54 at xhigh, with output around 190 tokens per second. From the official model card, against Claude Sonnet 5.5 and GPT-6.1 Terra:

CategoryBenchmarkHydrogenSonnet 5.5GPT-6.1 Terra
Agentic codingSWE-sweep13.5%3.1%5.8%
Agentic codingTerminal-Bench 4.157.0%63.9%—
Knowledge workAgents' Last Exam v1.171.6%60.4%69.8%
Knowledge workAutomationBench-AA76.1%71.8%60.2%
Knowledge workVending-Bench 2 (balance)$14,549$11,165$9,107
Computer useOSWorld-2.1 (offline)77.6%83.9%74.2%
ReasoningHLE-Diamond58.8%53.3%47.3%

Against its predecessor 3.8 Flash (18.2% on Terminal-Bench 4.1) that's more than a 3× jump. Other first-in-class results: Vals Tax Agent Bench (52.3%), Vals Legal Research Bench (55.8%, shared first place), AutomationBench-AA, and HLE-Diamond, plus #5 on the live Agent Arena leaderboard. On Plotly's DataAnalyticsBench it answered all 43 questions for $6.36 — the fourth model ever with a perfect score, ahead of Carbon (42/43) and Argon (41/43) at roughly half their cost.

Argon still owns the heaviest reasoning tasks — Chartography (71.6%), GraphWalks (84.2%), and SEC-bench Pro (83.4%).

Orchestrator in Antigravity

A new Orchestrator mode in Google Antigravity splits one session between the two models: Carbon handles planning while Hydrogen executes the routine steps. Google reports near-Carbon success rates at 24% lower quota spend and 19% faster completion.

Where to try Gemini 4 Hydrogen in Russia — no VPN needed

Hydrogen is already the default Google model on VEGA — pick Gemini 4 Hydrogen in the model selector, with all four thinking levels supported.


  • ru.vega.chat — AI model API service

  • officevega.com — the first office suite with AI

  • vega.codes — vibe coding and AI agents

VEGA API — all models·API docs·Представляем Gemini 4 Hydrogen

Read also

Qwen 4 Max: a 2.4T-parameter teammate for long-running work
Qwen 4 Max: a 2.4T-parameter teammate for long-running workNovember 16, 2026
Inkling 2 Small: Thinking Machines' Best Open Model Yet
Inkling 2 Small: Thinking Machines' Best Open Model YetNovember 13, 2026
MiniMax H3-Image: Open-Weight 4K Images in Under 10 Seconds
MiniMax H3-Image: Open-Weight 4K Images in Under 10 SecondsNovember 11, 2026

Comments · 0

VEGA - built in getmyai.io
PricingModelsBlogAbout CreatorTerms of ServicePrivacy PolicyMailing Policy