Gemini 4 Hydrogen, Google’s fast model for coding and agent work, is the quickest high-scoring model in its price class, but not the cheapest way to finish a task. Released on November 12, 2026 at $1.50 per million input tokens and $7.50 per million output tokens, it scores 51 on Artificial Analysis’s Intelligence Index at its default setting. GPT-6.1 Terra also reaches 51, at max effort, for roughly a third of the per-task cost of Hydrogen’s xhigh run.

What Google actually shipped

Google’s launch post, dated November 12, 2026 and updated November 13, describes Hydrogen as rolling out globally from that day. Developers get it in the Gemini API, Google AI Studio and Android Studio. Businesses get it through Gemini for business and Google Antigravity. Consumers see it in the Gemini app model dropdown and in Google Search. Pricing is flat across the full 1 million-token window, so there is no long-prompt surcharge.

Google’s own benchmark claims, reported here as vendor figures, include 57% on Terminal-Bench 4.1 against 18.2% for Gemini 3.8 Flash, and about 2.5 times the speed of 3.8 Flash at the high setting. Google also says Hydrogen surpasses Gemini 4 Argon on the Artificial Analysis index, 51 to 50.

Price per token is not price per task

The headline rate looks mid-range: above Claude Haiku 5.5 and GPT-6 Luna, below Claude Sonnet 5.5 and Gemini 4 Argon. But Artificial Analysis’s data shows Hydrogen is verbose. Its xhigh index run produced 304 million output tokens, against a class median of 77.3 million, and cost about $3.16 per index task. GPT-6.1 Terra, at a lower list price, cost about $0.96 per task at max effort for a score of 51 against Hydrogen’s 54 at xhigh.

In practice: if your workload is latency-bound, such as an interactive coding agent, Hydrogen’s roughly 171 output tokens per second at high effort is more than double Terra’s 69 at max. For a batch workload where total spend matters, run your own sample first. Effort levels are not comparable one-to-one across vendors, so treat every index figure here as directional.

The cheap tier is a different product

Claude Haiku 5.5 and GPT-6 Luna both list at $0.10/$0.50 per million tokens, 15 times below Hydrogen on input. On Artificial Analysis they score 36 and 28 at max effort. They fit classification, routing, extraction and other high-volume jobs where Hydrogen’s extra intelligence would be wasted. One catch on Haiku: Anthropic’s pricing page shows rates rising fivefold above 100,000 tokens, which erodes the advantage for long-context agents.

Mind the pricing calendar

Several rates here change within seven weeks. Gemini 4 Argon and Carbon introductory rates end November 30, 2026, after which Argon lists at $4.00/$20.00. Gemini 3.8 Flash is half Hydrogen’s price until December 31, 2026, then matches it from January 1, 2027. Hydrogen itself shows no introductory rate on Google’s page, so its price is the stable one. Budget on the post-promotion numbers.

How to choose

Pick Hydrogen for interactive coding and agent loops where speed and a high score matter and you are already on the Gemini API. Pick GPT-6.1 Terra when cost per completed task is the constraint and latency is not. Pick Claude Haiku 5.5 or GPT-6 Luna for high-volume simple work, and Claude Sonnet 5.5 when quality outranks spend. Skip Argon unless you have a reason to stay on it, and plan to move off Gemini 3.8 Flash before its price doubles.