Google releases Gemini 4 Hydrogen for coding and agentic tasks

Google has expanded its Gemini 4 family with Hydrogen, a model focused on speed for everyday coding and agentic tasks, available globally alongside existing reasoning-focused variants Carbon and Argon.

Orchestrated by Nadia Vale

November 12, 2026 at 1:37 PM EST2 min read

Key takeaways (3)

1.Hydrogen costs $1.50 per million input tokens and $7.50 output, below Carbon's $2 and $10 introductory prices.
2.Google says Orchestrator used 24% less quota on Terminal-Bench 4.0 with near-Carbon success; the post names no baseline.
3.Hydrogen offers four thinking levels, including a new xhigh tier, and up to 1 million output tokens.

Key takeaways (3)

1.Hydrogen costs $1.50 per million input tokens and $7.50 output, below Carbon's $2 and $10 introductory prices.
2.Google says Orchestrator used 24% less quota on Terminal-Bench 4.0 with near-Carbon success; the post names no baseline.
3.Hydrogen offers four thinking levels, including a new xhigh tier, and up to 1 million output tokens.

Google released Gemini 4 Hydrogen, a faster model for coding and agentic tasks, on November 12 and said it is rolling out globally.

Google positions Hydrogen as a cheaper execution tier that can run on its own or alongside its Gemini 4 Carbon and Gemini 4 Argon models. It is priced at $1.50 per million input tokens and $7.50 per million output tokens, flat across its 1 million-token context window, according to Google's announcement 1.

Automatica reported on November 10 3 that Google had made Carbon available, positioned above Argon for long-horizon coding, knowledge work, scientific reasoning and cybersecurity, at an introductory API price of $2 per million input tokens and $10 per million output tokens. Google's post describes Carbon and Argon as the models for hard, long-horizon problems. It says developers still need a fast model for writing code, running tests and verifying changes, and Hydrogen is meant for that role.

Access varies by surface and subscription. Hydrogen is available in Google AI Studio, the Gemini API, the paid tier in Android Studio, Gemini for business, the Gemini app and Google Search. Google Antigravity access is limited to paid Google AI Pro and Ultra subscribers. Gemini in Google Sheets begins with Pro and Ultra subscribers and eligible business subscriptions. Google's Search post 2 says the model gives Search more visual, structured responses to multi-part questions.

Google said Hydrogen scores 51 on the Artificial Analysis Intelligence Index at its default high thinking level, against Argon's 50, and 54 at the highest xhigh setting. For teams migrating from Gemini 3.8 Flash, Google said Hydrogen finishes tasks 2.5 times faster at the high setting and scores 57% on Terminal-Bench 4.1 versus 18.2%. Google also said unnamed partners, in early testing on graph, SQL and data-heavy enterprise workflows, saw runs 2x to 2.8x faster with 12% to 23% fewer tokens than 3.8 Flash.

The Terminal-Bench, speed, partner, safety and Orchestrator figures all come from Google's post, which includes no independent test results.

Google also introduced Orchestrator in Antigravity, which switches between models such as Carbon and Hydrogen within a session. On Terminal-Bench 4.0, Google said, Orchestrator reached near-Carbon task success rates while using 24% less quota and finishing 19% faster on average. The post does not state what those percentages are measured against.

On safety, Google said Hydrogen reduces overreach in agentic tasks by up to 81% compared with Gemini 3.8 Flash. It also reported a 96.7% honesty rate on its internal agent honesty evaluation, which covers impossible or blocked tasks. Google refers readers to a Gemini 4 Hydrogen model card for details; the card was not among the documents reviewed.

Hydrogen has four thinking levels: low, medium, high (the default) and a new xhigh tier. It accepts 1 million tokens of input and up to 1 million tokens of output.

Sources3 documents

1. blog.google

blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-4-hydrogen/

10 checked claims

2. blog.google

blog.google/products-and-platforms/products/search/gemini-4-hydrogen-search-upgrade/

1 checked claim

3. automaticapress.com

automaticapress.com/articles/google-makes-gemini-4-carbon-available-mwavumiu

1 checked claim

Citations link to the documents this article draws on; they do not imply independent corroboration.

llama.cpp releases v1.0.0, ties major version bumps to API breaks

Hugging Face has released llama.cpp v1.0.0, introducing stricter semantic versioning rules and significant updates to API, model support, security, and UI. The release aims to provide a more stable and predictable development environment for downstream applications.

Mira Sethi · about 5 hours ago

Google Research posts Harness² agent framework on arXiv and GitHub

The release matters because it targets a practical constraint in agent systems: performance depends not only on the underlying model but also on the prompts, tools, memory and orchestration around it.

Julian Park · about 5 hours ago

IBM Releases Two Granite 4.1 Preview Models Using LLMON Markup

The release addresses a structural weakness in how large language models process their own context.

Adrian Cole · about 20 hours ago