Skip to content
Neural Daily

ModelsNews

Gemini 4 Hydrogen: Argon-level smarts at $1.50 input

🇭🇺 Olvasd el magyarul →

Google released Gemini 4 Hydrogen, the fast, low-cost execution model of the Gemini 4 family. At its default setting it edges past Argon on the Artificial Analysis index, it is priced flat across the full 1M context, and it is already live in the API, the Gemini app and Search.

· 3 min read

A Gemini 4 Hydrogen hivatalos bejelentő grafikája kék háttéren
Image: Google · Press kit · Source

What changed

Carbon and Argon target long-horizon, deep reasoning. Hydrogen targets rapid execution: writing code, running tests and powering agent workflows (Google DeepMind). It also brings Gemini 4 to every tier, from free users to enterprise.

  • Artificial Analysis Intelligence Index: 51 at the default high thinking level versus 50 for Argon, and 54 at the new xhigh level.
  • Pricing: $1.50 per million input tokens and $7.50 per million output tokens, flat across the full 1M-token context window.
  • Four thinking levels: low, medium, high (default) and the new xhigh. Output can reach up to 1M tokens.
  • Compared with Gemini 3.8 Flash it finishes tasks 2.5x faster at high, scores 57% instead of 18.2% on Terminal-Bench 4.1, and cuts fabricated claims by more than 5x.

The numbers in detail

Google says Hydrogen natively batches file reads and searches in parallel when exploring large repositories instead of inspecting files one by one. In early partner tests it ran 2x to 2.8x faster than 3.8 Flash while using 12% to 23% fewer tokens.

  • HLE-Diamond: 58.8%, first in its class.
  • Vals Tax Agent Bench: 52.3%, first overall. Vals Legal Research Bench: 55.8%, tied for first.
  • Multimodal Knowledge Agent (MKA): 53.4%. Agents' Last Exam v1.1: 71.6%.
  • Agent Arena (arena.ai): fifth overall.

These are Google's own figures. The only independent number so far is the Artificial Analysis index.

Orchestrator: Carbon plans, Hydrogen executes

Google Antigravity gets Orchestrator, which switches models within a single session. On Terminal-Bench 4.0 it reached near-Carbon success rates while using 24% less quota on average and finishing 19% faster. For parallel agents, Carbon can coordinate while Hydrogen powers the worker and verifier agents (Gemini 4 Carbon: Google's new cyber-focused frontier model).

Safety

Google says Hydrogen reduces overreach in agentic tasks by up to 81% and cuts misaligned actions and task shortcutting by 12x across tool-use and coding evaluations. On impossible or blocked tasks it is up to 2x less likely to bypass permission boundaries. It is 2.3x more resilient on the Gray Swan indirect prompt injection benchmark, and false refusals on HealthBench Pro stay under 0.4%.

Where to get it

It is live in the Gemini API, AI Studio, Android Studio, Gemini for business, the Gemini app, and Google Sheets for Pro and Ultra subscribers. Paid AI Pro and Ultra subscribers can use it in Antigravity. Google Search now runs on Hydrogen for all users globally, with a broader query fan-out (Google).

Why it matters

If you run agents or coding workloads on 3.8 Flash, Hydrogen is the obvious upgrade: it outscores the previous flagship, Argon, at near-Flash prices (Gemini 4 Argon deep dive: Google's new frontier model). Flat pricing up to 1M tokens is especially useful for long-document and repo-scale work, since there is no long-prompt surcharge.

    By Mikola Máté Patrik

    Produced with the help of AI, reviewed by an editor. How we make our articles

    Reviewed and edited by the Neural Daily desk. About

    Related articles