For Developers/Models/Gemini 4 Hydrogen
Closed SourceGoogleReleased 2026-11

Gemini 4 Hydrogen

Google's fast Gemini 4 model for everyday coding and agent work, with a free API tier

Context window

1M (shared input + output)

Input / 1M tokens

$1.50

Output / 1M tokens

$7.50

Provider

Google

Paid tier: $1.50 input / $7.50 output per 1M tokens (output includes thinking tokens), flat across the full 1M context, with no introductory price. Cached input $0.075 (implicit caching only; explicit caching is not supported). Batch and Flex: $0.75 / $3.75 (cached $0.0375). Priority: $2.70 / $13.50. A free tier is available with rate limits (free-tier content is used to improve Google products). Grounding with Google Search: 5,000 free requests a month shared across Gemini 3 and newer models, then $14 per 1,000. · Data verified 2026-11-17

Gemini 4 Hydrogen (November 12, 2026) is the fast, efficient model in Google's Gemini 4 family, built for everyday coding and agentic workflows alongside the frontier models Gemini 4 Carbon and Gemini 4 Argon. It is adapted from Gemini 4 Carbon, accepts text, image, video, audio and PDF input in a 1,048,576-token window shared by input and output, and writes up to 65,536 tokens per synchronous response or up to 1M with long decode. It costs $1.50/$7.50 per 1M tokens with a free API tier, and Google calls it the recommended upgrade from Gemini 3.8 Flash. Google reports it finishes tasks 2.5x faster than 3.8 Flash at its default high setting, scores 57.0% on Terminal-Bench 4.1 versus 18.2% for 3.8 Flash, and edges Gemini 4 Argon on the Artificial Analysis Intelligence Index (51 vs 50 at high, 54 at xhigh). It also runs in Google Search for all users, the Gemini app and Google Antigravity.

How to access it

Generally available since November 12, 2026 in the Gemini API (model ID gemini-4-hydrogen) on the free and paid tiers, in Google AI Studio and Android Studio, and listed on Gemini Enterprise Agent Platform. It is the default model for the Antigravity agent in Gemini Managed Agents. Consumers get it in Google Search (all users, globally), in the Gemini app model picker, and through auto model selection for free Gemini app users. Paid Google AI Pro and Ultra subscribers can use it in Google Antigravity, and Gemini in Google Sheets gets it starting with Pro and Ultra. Businesses get it in Gemini for business through the API or model picker, Antigravity and Sheets.

Strengths

  • ✓57.0% on Terminal-Bench 4.1 (cheating-adjusted), up from 18.2% for Gemini 3.8 Flash (Google-reported)
  • ✓Artificial Analysis Intelligence Index v5.0.1: 51 at the default high level (Gemini 4 Argon 50) and 54 at xhigh
  • ✓About 2.5x faster than Gemini 3.8 Flash at its default setting, with more than 5x fewer fabricated claims (Google-reported)
  • ✓Free API tier and flat $1.50/$7.50 pricing across the full 1M context
  • ✓Four thinking levels (low, medium, high by default, xhigh) and up to 1M output tokens with long decode
  • ✓Ahead of Claude Sonnet 5.5 and GPT-6.1 Terra on Agents' Last Exam v1.1, AutomationBench-AA, Vending-Bench 2 and HLE-Diamond in Google's table

Best for developers who...

Everyday coding, interactive web apps and tool-using agentsParallel subagent work, including pairing with Gemini 4 Carbon for planningMultimodal document workflows over images, video, audio and PDFsTeams moving off Gemini 3.8 Flash or older Gemini 3 and 2.5 models

When to choose it (and when not to)

Reach for Gemini 4 Hydrogen when...

  • →Everyday coding and agent work where Gemini 4 Carbon or Argon would be slower or cost more ($4/$20 from December 1, 2026)
  • →You are upgrading from Gemini 3.8 Flash, which Google now lists as its previous-generation Flash model
  • →You want a Google model on the free API tier with a 1M-token context and long outputs
  • →Multi-model setups where Carbon plans and Hydrogen runs code edits, subagents and tests

Look elsewhere if...

  • ✕The hardest long-horizon reasoning, where Google points to Gemini 4 Carbon
  • ✕Code that relies on explicit context caching, temperature/top_p/top_k or the minimal thinking level, which Hydrogen does not support
  • ✕Lowest cost per token: Gemini 3.8 Flash costs $0.75/$3.75 until December 31, 2026, and Gemini 3.5 Flash-Lite $0.30/$2.50
  • ✕Computer use and terminal work where Google's own table puts Claude Sonnet 5.5 ahead (OSWorld-2.1, Terminal-Bench 4.1)

How to use it

  • ›Thinking defaults to high; test medium for speed and keep xhigh for hard problems
  • ›Remove temperature, top_p and top_k from requests; out-of-range values return a 400 error
  • ›Raise hardcoded max_output_tokens caps to 65,536 and use long decode (background execution) for longer outputs
  • ›Use previous_interaction_id in the Interactions API to keep conversation state and thought signatures across turns

Quickstart

Python
from google import genai

client = genai.Client()

interaction = client.interactions.create(
    model="gemini-4-hydrogen",
    input="Build an interactive single-file HTML token bucket rate-limiter simulator with live sliders for burst size and request traffic."
)

print(interaction.output_text)

Install with pip install -U google-genai. Sample from Google's What's new in Gemini 4 Hydrogen guide (Interactions API); works on the free tier.

API model id: gemini-4-hydrogen

Benchmarks

BenchmarkScoreNotes
Terminal-Bench 4.157.0% (high)Google model card, cheating-adjusted, mini-swe-agent vision harness, average of 5 attempts per task; Claude Sonnet 5.5 63.9%, Gemini 3.8 Flash 18.2%
Artificial Analysis Intelligence Index v5.0.154 (xhigh)51 at the default high level vs Gemini 4 Argon 50, per Google; Artificial Analysis lists the same scores
SWE-sweep13.5%Google model card, Antigravity CLI harness, subtask-weighted reward; Claude Sonnet 5.5 3.1%, GPT-6.1 Terra 5.8%
Agents' Last Exam v1.171.6%Google model card; Claude Sonnet 5.5 60.4%, GPT-6.1 Terra 69.8%
AutomationBench-AA76.1% (xhigh)Google model card, Artificial Analysis leaderboard; Claude Sonnet 5.5 71.8%, GPT-6.1 Terra 60.2%
Vending-Bench 2$14,549Final bank balance, Andon Labs leaderboard per Google; Claude Sonnet 5.5 $11,165, GPT-6.1 Terra $9,107
OSWorld-2.1 offline subset77.6%Google model card, 82-task offline subset, best of 3 runs; Claude Sonnet 5.5 83.9%, GPT-6.1 Terra 74.2%
MKA (Multimodal Knowledge Agent)53.4%Google model card; Claude Sonnet 5.5 52.0%, GPT-6.1 Terra 43.8%
Chartography66.6%Google model card, no tools; Claude Sonnet 5.5 61.6%, GPT-6.1 Terra 63.2%
Terminal-Bench-Science 0.151.4%Google model card; Claude Sonnet 5.5 59.9%, GPT-6.1 Terra 58.3%
HLE-Diamond58.8% (xhigh)Google model card, Artificial Analysis leaderboard; Claude Sonnet 5.5 53.3%, GPT-6.1 Terra 47.3%

Source: Gemini 4 Hydrogen model card

Tools powered by Gemini 4 Hydrogen

Compare Gemini 4 Hydrogen

Compare Gemini 4 Hydrogen with any other model

Build a comparison →
All model comparisons →

Learn the concepts