Google has released Gemini 4 Hydrogen, a fast AI model, priced below its bigger Gemini 4 siblings, that replaces Gemini 3.8 Flash and, unlike the two models before it, is available to developers on a free tier as well as a paid one. Announced on 12 November (US time), it is rolling out in the Gemini API, Google AI Studio, the Gemini app, Google Search and Google’s Antigravity coding tool, according to Google’s announcement.
Google says Hydrogen is both faster and smarter than the Flash model it replaces, and that it beats the larger Gemini 4 Argon on one widely watched ranking. Independent testers agree on some of that and not on all of it. Here’s who gets Hydrogen, what it costs in Australian dollars, and which claims hold up.
What Gemini 4 Hydrogen is, and who can use it
Hydrogen is the third model in the Gemini 4 family, after Gemini 4 Argon (October) and Gemini 4 Carbon, which we covered on Wednesday. Google’s model card says Hydrogen is adapted from Carbon, takes text, images, audio and video, and has a 1 million-token context window, with 64,000 output tokens by default and up to 1 million through a background “long decode” mode. It has four thinking levels: low, medium, high (the default) and a new xhigh.
Google’s stated availability, which it says is global:
- Developers: the Gemini API in AI Studio and Android Studio. Google’s developer guide says Hydrogen is on both the free and paid tiers.
- Paid Google AI Pro and Ultra subscribers: Google Antigravity, its coding tool, where Google also introduced an “Orchestrator” that switches between Carbon and Hydrogen in one session.
- Businesses: Gemini for business, plus Gemini in Google Sheets for eligible business subscriptions.
- Everyone: Google Search, where the company says Hydrogen is available globally to all users. In the Gemini app it says Hydrogen is in the model menu, with a footnote that free users get it through automatic model selection “when it’s the appropriate model for the prompt”.
That footnote matters. Ars Technica reported that AI Plus and free Gemini app users are still on an older Flash-Lite model, while The Decoder described Hydrogen as available to everyone in the app. Google’s wording sits between the two: free users can get Hydrogen, but not necessarily on every prompt. We haven’t been able to confirm how often that happens.
What it costs, in Australian dollars
Google lists Hydrogen at US$1.50 per million input tokens and US$7.50 per million output tokens, flat across the whole 1 million-token window, with cached input at US$0.075, on its Gemini API pricing page. At the Reserve Bank of Australia’s 12 November rate of US$0.6841 per Australian dollar (RBA), our straight conversion is about A$2.19 for input and A$10.96 for output per million tokens, or roughly A$0.11 for cached input. That is a currency conversion, not Google’s Australian billing price, and it excludes GST.
Ars wrote that this is “unchanged from 3.8 Flash”. That is true only against Flash’s standard rate. Google’s Gemini 3.8 Flash guide says Flash is on an introductory price of US$0.75 and US$3.75 until 31 December 2026, rising to US$1.50 and US$7.50 on 1 January 2027. So until the end of the year, Hydrogen costs twice what Flash does per token (about A$1.10 and A$5.48 for Flash at the same RBA rate).
It is cheaper than the bigger Gemini 4 models, though. Google’s pricing page lists Carbon at US$2 and US$10 through 30 November, rising to US$4 and US$20 from 1 December (as in our earlier Carbon article), and Ars reports the same pattern for Argon. Developers on the free tier pay nothing, but Google’s pricing page says free-tier content is used to improve its products, while paid-tier content is not.
What Google claims, and what independent tests show
Google’s headline claims are that Hydrogen finishes tasks 2.5 times faster than Gemini 3.8 Flash at its default setting, more than triples its Terminal-Bench 4.1 score (57% against 18.2%), cuts fabricated claims by more than five times, and beats Argon on the Artificial Analysis Intelligence Index (51 against 50 at high, 54 at xhigh). Here is how they compare with outside sources:
- Speed and the Argon comparison largely hold up. Artificial Analysis’s Google model table lists Hydrogen (High) at 51, Argon (High) at 50, Carbon (High) at 60 and Gemini 3.8 Flash (High) at 31. It shows Hydrogen (High) taking 6.0 minutes per index task against 14.8 for Flash, which works out to about 2.5 times faster by our arithmetic, and 30.2 minutes for Argon.
- The top score costs more thinking. Artificial Analysis’s xhigh page gives Hydrogen 54 at xhigh, but calls it “very verbose”, with 304 million output tokens across the index against 72.8 million for the median model, and US$3.16 per task. At the low setting it scores 27 at US$0.29 per task. The Decoder reports US$1.87 for high.
- Rival comparisons are mixed. The Decoder reports Artificial Analysis scoring Hydrogen just behind Claude Sonnet 5.5 and GPT-6.1 Sol at 55 each at their top settings, and weaker on terminal-style agent work (54% on Terminal-Bench 4.0, against 64% for Claude Sonnet 5.5). It also reports a hallucination rate of 8% on Artificial Analysis’s AA-Omniscience test.
- One Google claim doesn’t match. Google says Hydrogen ties for first on Vals’ Legal Research Bench with 55.8%. Vals’ own model page shows 53.85%, ranked sixth of 94. It does confirm first place on Tax Agent Bench (52.30%) and puts Hydrogen sixth of 37 on its overall Vals Index (68.49%, behind Claude Fable 5.2). Vals evaluated it at xhigh, so the difference may reflect settings or versions, and we can’t tell which.
- Safety and honesty figures (for example, a 96.7% score on Google’s own agent honesty evaluation, and up to 81% less overreach in agentic tasks) come from Google’s internal tests. We haven’t seen independent checks, and the Antigravity Orchestrator figures (24% less quota, 19% faster on Terminal-Bench 4.0) are likewise Google’s own.
What changes for Australian users
In Search, Google says Hydrogen is better at long, multi-part questions and typos, returns more tables and images, and fans out to a wider range of links, though Ars notes it has “heard that from Google before”. Because Hydrogen replaces 3.8 Flash across Google’s products, as Ars reports, most people will notice it as faster, more visual answers rather than a new model name. Google says its Search upgrade covers all languages supported in AI Mode.
For developers, migration is mostly a change of model ID, but Google’s guide warns that Hydrogen shares a single 1,048,576-token window across input, thinking and output, and recommends raising any hard-coded low output caps to 65,536. The guide also calls Hydrogen the “recommended upgrade” from Gemini 3.8 Flash and earlier Gemini 3 and 2.5 models.
Our read
Our read: Hydrogen looks like the most practical Gemini 4 model for most people, because the evidence for its speed and its mid-table intelligence ranking comes from Artificial Analysis as well as Google, and it costs less than Carbon and Argon. Two things deserve care. First, the per-token price is double Flash’s current introductory price, and the xhigh setting burns far more tokens, so your real bill depends on which thinking level you choose. Second, “beats Argon” is true on one index and weaker on terminal and legal benchmarks, so test it on your own work before moving anything important over. We haven’t run our own tests of Hydrogen; the figures above are Google’s or third parties’.
What we couldn’t confirm: how often free Gemini app users actually get Hydrogen, whether Google will list Australian-dollar pricing for the Gemini API, and whether any independent test reproduces Google’s safety numbers.
Featured image: Google’s headquarters in Mountain View, California, photographed in 2016. It is a stock view of the company’s campus, not an image of Gemini 4 Hydrogen. Photo: Asoundd via Wikimedia Commons, licensed under CC BY-SA 4.0; cropped to 3:2, resized to 1200×800 and converted to PNG. The adapted image is shared under the same licence. Google trademarks are shown for editorial purposes and do not imply endorsement.
.jpg){kind=link}