Google logo

Gemini 4 Hydrogen

Multimodal
Google

Gemini 4 Hydrogen is Google DeepMind's Gemini 4 model for everyday coding and agentic workflows, announced and generally available on 2026-11-12 (model id gemini-4-hydrogen). It is adapted from Gemini 4 Carbon and is available on the Gemini API free and paid tiers, Google AI Studio, Vertex AI, the Gemini app, Google Search, and Google Antigravity. It uses a 1,048,576-token unified context window with up to 64K default synchronous output tokens and long decode up to 1M tokens, and adds a fourth thinking level, xhigh, alongside low, medium, and high (default high). Gemini API paid pricing is flat across the 1M window at $1.50 per million input tokens, $7.50 per million output tokens (including thinking tokens), and $0.075 per million implicitly cached input tokens; explicit caching is not supported. Vertex AI global endpoints use the same $1.50 / $7.50 / $0.075 rates; non-global endpoints are billed at a 10% premium. The knowledge cutoff is March 2026.

Key Specifications

Parameters
-
Context
1.0M
Release Date
November 12, 2026
Average Score
66.8%

Timeline

Key dates in the model's history
Announcement
November 12, 2026
Last Update
November 18, 2026
Today
November 20, 2026

Technical Specifications

Parameters
-
Training Tokens
-
Knowledge Cutoff
-
Family
-
Capabilities
MultimodalZeroEval

Pricing & Availability

Input (per 1M tokens)
$1.50
Output (per 1M tokens)
$7.50
Max Input Tokens
1.0M
Max Output Tokens
1.0M
Supported Features
Function CallingStructured OutputCode ExecutionWeb SearchBatch InferenceFine-tuning

Benchmark Results

Model performance metrics across various tests and benchmarks

Other Tests

Specialized benchmarks
Agents' Last Exam
Agents' Last Exam v1.1 score (122-task v1.1 set, Oct 17, 2026 task version, without the Oct 26, 2026 update). Official methodology: self-computed with the Antigravity-CLI (agy) version 1.4.8 harness through official integration and a 5-hour window, with safety filters enabled; responses flagged by safety filters are returned as empty strings, though the episode can continue. Averaged rubric Score. • Self-reported
71.6%
Chartography (no tools)
Self-computed without tools, following the official Surge public leaderboard, as stated in the official Gemini 4 Hydrogen evaluation methodology. • Self-reported
66.6%
OSWorld 2.1 (partial)
OSWorld 2.1 offline (no-internet) subset (82 tasks), partial score; self-computed, maxed over 3 runs with a single attempt per run; official GitHub OSWorld docker and evaluator, default 1080p, max step length 500; updated Gemini CUA harness with parallel batch tool calling, all screenshots in history, and a compaction procedure following the same setup as reported in the Opus 5.5 system card. Official methodology: safety-filter flagged responses are returned as empty strings (episode can continue); PyAutoGUI actuation with screenshot-only observation space and UI-specific function declarations. • Self-reported
77.6%
Terminal-Bench-Science 0.1
Terminal-Bench Science 0.1; self-computed using a mini-swe-agent vision harness, averaged over 5 attempts per task (Average Reward), as stated in the official Gemini 4 Hydrogen evaluation methodology. • Self-reported
51.4%

License & Metadata

License
proprietary
Announcement Date
November 12, 2026
Last Updated
November 18, 2026

Compare Gemini 4 Hydrogen

All comparisons

Similar Models

All Models

Recommendations are based on similarity of characteristics: developer organization, multimodality, parameter size, and benchmark performance. Choose a model to compare or go to the full catalog to browse all available AI models.