Google Launches Gemini 4 Hydrogen Across AI Products

By James Calloway

Google says Gemini 4 Hydrogen completes tasks 2.5 times faster at its high default setting than Gemini 3.8 Flash, while accepting 1 million input tokens and producing as many as 1 million output tokens. The company says the launch brings “cutting-edge intelligence to everyone.”

Google announced the model after Carbon became available this week and chose Hydrogen rather than keeping the “Flash” branding.

What changed from Gemini 3.8 Flash?

Google describes Hydrogen as a major standalone upgrade over Gemini 3.8 Flash, which launched in September, with significant gains in speed and intelligence. The company reports these comparisons:

• Hydrogen more than triples Terminal-Bench 4.1 scores, reaching 57% against 18.2% for Gemini 3.8 Flash.
• It cuts fabricated claims by more than 5x.
• It adapts cleanly when a tool call fails or a search returns no results.

In practical terms, Google is selling Hydrogen as a faster, larger and more reliable successor to Flash.

How do the benchmark claims stack up?\n

Google says Hydrogen scores 51 on the Artificial Analysis Intelligence Index at its default high thinking level, ahead of Argon at 50. The score reaches 54 at the highest xhigh setting. Google also says the model often completes tasks faster than other models in its class.

The launch includes several external benchmark comparisons:

• On Agent Arena, arena.ai’s live benchmark for deep research, data analysis and broad codebase navigation, Hydrogen ranks fifth.
• For deep domain reasoning, Hydrogen leads its class on HLE-Diamond with 58.8%, ranks first overall on Vals Tax Agent Bench with 52.3%, ties for first on Vals Legal Research Bench with 55.8%, and leads its class on AutomationBench-AA.
• It leads other models on the new Multimodal Knowledge Agent benchmark with 53.4% for agentic knowledge work over complex visual files. It also records 71.6% on Agents’ Last Exam v1.1 for end-to-end professional workflows.

Google says partners have run Hydrogen 2x to 2.8x faster than Gemini 3.8 Flash while reducing token usage by 12% to 23%.

Compared with Gemini 3.8 Flash, Google reports that Hydrogen cuts overreach in agentic tasks by up to 81% and under-service by up to 79%. It also cuts misaligned actions and task shortcutting across tool-use and coding evaluations by 12x.

For impossible or blocked tasks, Google says Hydrogen is up to 2x less likely to bypass permission boundaries. It also reaches a 96.7% honesty rate on Google’s internal agent honesty evaluation.

The release gives buyers a broad view of speed, reasoning and agent performance, with the 96.7% honesty result coming from Google’s own internal evaluation.

What changes in the Gemini app and Search?

Google says Hydrogen better follows and sticks to a user’s tone, format and rules, and returns clear tables and visual breakdowns. The company says email drafts, study guides, rewrites and documents are ready to use on the first try. It also more frequently calls the right tools, including Connected Apps, and skills for the job.

Google’s Gemini model picker presents five routes:

• Quick answers: 3.5 Flash-Lite
• Everyday help: 4 Hydrogen
• Detailed research: 3.1 Pro on AI Pro
• Code and analysis: 4 Argon
• Most complex work: 4 Carbon

Google says free Gemini app users can access Hydrogen through auto model selection when the system considers it appropriate for the prompt.

In Search’s AI Mode, Google says Hydrogen better understands intent. Users can ask questions in a natural flow of thought, including multi-part questions, hard-to-understand typos and highly specific requests. Responses become more visual and include a wider range of relevant links.

For everyday users, Hydrogen raises the app’s default capability while keeping specialist models available for harder work.

How does Antigravity split work between models?

Google says 4 Hydrogen adds an “Extra” thinking tier to Antigravity’s existing Low, Medium and High tiers. Orchestrator dynamically switches between models such as Carbon and Hydrogen within the same session instead of using one model from start to finish.

Google says: “Carbon that may offload routine steps to Hydrogen. Same Carbon-level quality, fewer tokens used.”

For workflows requiring parallel agents, Carbon can coordinate the work while Hydrogen powers worker and verifier agents. Those agents can build and test changes at the same time across isolated workspaces. Google says the arrangement helps teams ship complex projects faster without increasing token costs.

Antigravity’s pitch is division of labor: Carbon directs the work, while Hydrogen executes and checks it.

What does Hydrogen cost, and where is it available?\n

For developers, Google says Hydrogen costs $1.50 per million input tokens and $7.50 per million output tokens.

Google says the model is available across all tiers of Google AI Studio, the Gemini API, Gemini for business, the Gemini app and Google Search. Paid Google AI Pro and Ultra subscribers can use it in Google Antigravity.

Access begins broadly across Google’s developer and consumer products, though Antigravity remains tied to the paid AI Pro and Ultra tiers. The next test is practical: whether Google’s claimed 2x-to-2.8x partner speed gains and 12%-23% token savings hold in production.