Google launched Gemini 4 Hydrogen on November 12 for coding and agentic workflows. It is available through Google AI Studio and the Gemini API, with app, Search and business surfaces also announced. Antigravity access is specified for paid AI Pro and Ultra subscribers.

An execution tier alongside Carbon and Argon

Hydrogen supports low, medium, high and xhigh thinking levels. High is the default. As checked November 13, Artificial Analysis lists High at 51 on Intelligence Index v5.0.1 and the separate Xhigh configuration at 54; Argon High is 50. These are effort-qualified snapshots, not proof of leadership on every task.

Google’s standalone 57% Terminal-Bench claim is for version 4.1. Its Orchestrator result uses a multi-model workflow on version 4.0. Neither is substituted for an independently verified standalone Hydrogen TB4.0 harness result.

Limits and costs still matter

The launch lists $1.50 per million input tokens and $7.50 per million output tokens across its stated one-million-token window. The model card defines a combined input-output context: default synchronous output is up to 64K, while output up to 1M requires background long decode. These rounded labels alone do not establish exact integer limits.

Google describes improved execution speed and fewer fabricated claims, but those are vendor comparisons under particular conditions. Its model card says the full safety report is forthcoming; broad availability does not remove the need to verify consequential outputs.