Artificial Analysis Scores Google's Gemini 4 Hydrogen 51 Against Argon's 50 at High Thinking, With $1.87 per Index Task Versus $4.97
Google's Gemini 4 Hydrogen costs $1.50 and $7.50 per million tokens. Artificial Analysis scores it 51 at high thinking against Argon's 50; Google Cloud lists 3.8 Flash at half that price through December 31.
Editor's Note ·
- Correction:
- The article says Google Cloud's Hydrogen page calls the model "a short-term generally available (GA) release subject to a minimum 45-day retirement notice policy". That wording is not on the Hydrogen page (docs.cloud.google.com/vertex-ai/generative-ai/docs/models/gemini/4-hydrogen, read November 13, 2026), which lists only "Launch stage: GA" and "Release date: November 12, 2026". It paraphrases Google Cloud's model-versions page (cloud.google.com/vertex-ai/generative-ai/docs/learn/model-versions), which lists gemini-4-hydrogen among "Models available for shorter availability periods" and says that when Google Cloud schedules such a model for retirement it posts a fixed date "that gives you at least 45 days to migrate".
- Clarification:
- The article reports the Google Cloud pricing page's $0.75 and $3.75 for 3.8 Flash through December 31, 2026 and $2.00 and $10.00 for Argon through November 30, 2026 without saying how they are delivered. The page's footnote reads "Promotional pricing provided through 50% credits back on net spend on select models within a given period", so these figures are promotional prices provided through credits, not a lower standard rate.
Overview
Google released Gemini 4 Hydrogen on November 12, 2026. Google’s announcement says it is available in all tiers of Google AI Studio and the Gemini API and is “priced flat across its entire 1 million-token context window at $1.50 per million input tokens and $7.50 per million output tokens”. Ars Technica reports that the model has replaced Gemini 3.8 Flash across Google’s products.
Artificial Analysis (AA), a third-party benchmarking site, scores Hydrogen 51 on its Intelligence Index at high thinking, the default, against 50 for Gemini 4 Argon at high. AA’s cost per index task is $1.87 for Hydrogen at high and $4.97 for Argon at high. At the new xhigh level, AA scores Hydrogen 54.
The cited sources differ on one price. Ars says Hydrogen’s price is “unchanged from 3.8 Flash”, while AA lists 3.8 Flash at $0.75 and $3.75. A Google Cloud pricing page explains both (see “Pricing”).
What Google reports
Google’s post lists four thinking levels, low, medium, high (the default) and xhigh. It says Hydrogen “surpasses Argon on the Artificial Analysis Intelligence Index at its default high thinking level (51 vs. 50, and reaching 54 at its highest xhigh thinking setting)”. Relative to 3.8 Flash, Google says Hydrogen finishes tasks 2.5 times faster at its high default setting and scores 57% against 18.2% on Terminal-Bench 4.1. Google’s evaluation methodology says the Hydrogen figure is self-computed with a mini-swe-agent vision harness at high effort level, and that “All the results for non-Gemini models are sourced from providers’ self-reported numbers unless otherwise mentioned below”.
The post also says Hydrogen “matches or outperforms competing models in its class across core developer and enterprise workflows”. The post and the model card do not define the class; an Ars Technica caption for a Google-credited chart reads “Hydrogen compared to the mid-tier options from OpenAI and Anthropic.”
The model card compares Hydrogen with Claude Sonnet 5.5 and GPT-6.1 Terra in ten benchmark rows. By the Herald’s count, the Sonnet 5.5 column is higher on three: Terminal-Bench 4.1 (63.9% against Hydrogen’s 57.0%), OSWorld-2.1 on the offline subset (83.9% against 77.6%) and Terminal-Bench-Science 0.1 (59.9% against 51.4%). The GPT-6.1 Terra column is higher on one, Terminal-Bench-Science 0.1 (58.3%). The card does not say whether these two columns are the class the post means.
What Artificial Analysis measured
AA’s model pages, read on November 13, 2026, give these figures for Intelligence Index v5.0.1. Cost and time are per index task as AA defines them.
| Model and setting | Index score | Cost per task | Time per task |
|---|---|---|---|
| Hydrogen, low | 27 | $0.29 | 1.3 min |
| Hydrogen, medium | 41 | $0.80 | 3.7 min |
| Hydrogen, high | 51 | $1.87 | 7.9 min |
| Hydrogen, xhigh | 54 | $3.16 | 13.7 min |
| Argon, high | 50 | $4.97 | 30.0 min |
| 3.8 Flash, medium | 30 | $6.26 | 13.0 min |
| 3.8 Flash, high | 31 | $7.77 | 14.7 min |
On those pages Hydrogen at high ranks 21st of 78 on intelligence, Argon at high 22nd of 78 and Hydrogen at xhigh 16th of 78. The pages list $1.50 and $7.50 per million tokens for Hydrogen, $2.00 and $10.00 for Argon and $0.75 and $3.75 for 3.8 Flash, so 3.8 Flash’s cost per task is higher than Hydrogen’s despite the lower per-token price.
AA’s summaries describe Hydrogen at xhigh as “slower than average per task and very verbose”, at high as “quicker than average per task, however somewhat verbose” and at medium as “notably quick per task and fairly concise”. Google’s post says Hydrogen is “often finishing tasks faster and using fewer output tokens” across thinking levels, while its model card says “at times the model may use more tokens to maximize performance, especially at higher thinking levels”.
Google’s post does not say which 3.8 Flash setting or tasks its 2.5 times refers to. AA, which computes time per task from output tokens and output speed, lists 7.9 minutes for Hydrogen at high against 14.7 for 3.8 Flash at high and 13.0 at medium, which a Google Cloud page gives as 3.8 Flash’s default; the Herald’s arithmetic gives about 1.9 and 1.6 times. The cited pages do not say whether the two measures match.
AA is not wholly separate from Google’s figures: Google’s methodology sources its AutomationBench and HLE-Diamond results from the AA leaderboard, and its Carbon announcement says Google partnered with AA on the MKA benchmark.
Pricing
Google Cloud’s pricing page, read on November 13, 2026, says 3.8 Flash is offered “with introductory pricing of $0.75 / $3.75 per 1M tokens input / output through December 31, 2026”, with $1.5 and $7.5 from January 1, 2027. So Hydrogen’s $1.50 and $7.50 is twice 3.8 Flash’s price through December 31 and equal afterwards (the Herald’s comparison), which fits Ars’s “unchanged”. That page covers Google Cloud’s platform, not necessarily the API Ars and AA describe.
Ars also reports that Argon and Carbon, “currently $2/$10”, will “go up to $4/$20 starting next month”. The Google Cloud page lists Argon at $2.00 and $10.00 through November 30, 2026 and $4.00 and $20.00 from December 1, 2026. From December 1, Argon’s listed price would be about 2.7 times Hydrogen’s on both input and output (the Herald’s arithmetic from $4.00 and $1.50, and $20.00 and $7.50). The Herald’s earlier Argon coverage described the $2 and $10 introductory price. Google’s Carbon post, published November 10, said Argon “remains the ideal choice for the majority of software engineering and knowledge work”; The Decoder writes that Hydrogen “should make that advice moot”.
Developer-facing details
Google Cloud’s Hydrogen page calls the model “a short-term generally available (GA) release subject to a minimum 45-day retirement notice policy”. Among other items, it lists thinking levels LOW, MEDIUM, HIGH (default) and XHIGH; setting thinking_budget or a MINIMAL level returns an HTTP 400 error, custom temperature, top_p and top_k values are ignored, and only implicit caching is supported. The 3.8 Flash page lists MEDIUM as its default, so by the Herald’s comparison, a request on Google Cloud’s platform that sets no thinking level runs at high on Hydrogen where it ran at medium on 3.8 Flash.
Dated changes on those Google Cloud pages, in order: Argon and Carbon prices change on December 1, 2026; 3.8 Flash’s introductory price ends after December 31, 2026; and from January 12, 2027, Hydrogen supports cache control with TTL configuration, with cache writes billed at 1.25 times the standard input price by default.
Open points
The model card says the full Safety & Security Report “will be available in the coming days”. On the Gemini app, Google’s post says free users get Hydrogen “via auto model selection when it’s the appropriate model for the prompt”, and 9to5Google reports the same. Ars writes that “AI Plus and free users are still stuck with the old 3.5 Flash-Lite model”. None of the cited pages says how often auto selection chooses Hydrogen.
The Herald’s contributor model is built by Anthropic, whose Claude Sonnet 5.5 is one of the comparison columns in Google’s model card.