Google's new everyday AI model beats a premium sibling, at double its predecessor's current price

Google released Gemini 4 Hydrogen on November 12 as the new everyday model across its developer tools, workplace apps and Search. It is meant to replace Gemini 3.8 Flash, Google's previous mid-priced model for routine automated work.

The headline for buyers is the price. Hydrogen costs $1.50 per million tokens of input and $7.50 per million of output. Tokens are chunks of text, roughly three-quarters of a word each. That is double what Gemini 3.8 Flash costs today.

Flash will not stay cheaper for long. Its current rate of $0.75 in and $3.75 out runs only through December 31, according to Google's published API price list. On January 1 it rises to exactly Hydrogen's price. For companies that built on Flash, Google's mid-range tier is effectively doubling.

What they get for the money is a much more capable model. On the Artificial Analysis Intelligence Index, an independent composite of reasoning, coding and knowledge tests, Hydrogen scores 51 at its default setting. Gemini 3.8 Flash scores 31.

Cheaper than Google's premium line, and better than part of it

Hydrogen also edges past Gemini 4 Argon, one of Google's two premium models, which scores 50 on the same index. Argon launched less than four weeks ago. The Decoder noted that Google had recommended Argon for daily coding just two days before Hydrogen arrived.

Argon and its sibling Carbon list at $2 in and $10 out, rising to $4 and $20 on December 1. Developer Simon Willison notes Hydrogen is only about 25% cheaper than both today. After the increase, they will cost more than two and a half times as much as Hydrogen per token.

Hydrogen has four effort settings, from low to a new "xhigh," that trade cost for quality. At xhigh it reaches 54 on the index. Artificial Analysis puts the cost of an average test task at $3.16 there, against $1.87 at the default setting.

The Decoder reports that xhigh cost is roughly one-third of Anthropic's Claude Sonnet 5.5 at its maximum setting, which scores 55. Quality falls fast at the cheap end, though. Hydrogen scores 41 at medium effort and 27 at low.

On the same index, Google's premium Carbon model still scores 59, well above Hydrogen. Willison's informal side-by-side test found the quality gap with Carbon "quite stark." Yet his Hydrogen run cost about 4 cents, against roughly 20 cents for Carbon.

What Google claims, and what outsiders found

Most of the strongest claims come from Google. It says Hydrogen finishes tasks 2.5 times faster than Flash and makes up facts more than five times less often. It also cites first place on Vals AI's tax-agent test and a tie for first on its legal-research test.

Some independent evidence supports the efficiency story. Artificial Analysis's figures show Hydrogen used about half as many input tokens as Flash to complete its test suite. Google says early partners saw 12% to 23% fewer tokens on data-heavy work, but it named no partners.

Fewer tokens can partly offset the higher rate, because companies pay for every token a model reads and writes. Whether a given workload gets cheaper or dearer will depend on how much of it Hydrogen can finish in fewer steps.

Outside testing also found limits. Artificial Analysis found Hydrogen fabricates answers about 8% of the time on its accuracy test, according to The Decoder. It also trails Claude Sonnet 5.5 on tasks run directly in a computer's command line, scoring 54% to Sonnet's 64% on one standard test.

The low end of the market moves up

Hydrogen is not a bargain model in the way some rivals' smallest offerings are. Willison notes Anthropic's Claude Haiku 5.5 and OpenAI's GPT-6 Luna both list at $0.10 in and $0.50 out. That makes Hydrogen about 15 times pricier per token.

Google still sells cheaper tiers. Gemini 3.5 Flash-Lite lists at $0.30 in and $2.50 out, with no scheduled increase. But the model Google now markets as its everyday workhorse sits well above that floor.

Consumers see a similar reshuffle. Notebookcheck reports Hydrogen is now the headline model in the $4.99-a-month Google AI Plus plan. Google emailed Plus subscribers that they would lose its Pro models on November 13, though its help page still listed Pro under the plan. Free users get Hydrogen only when Google's automatic model picker chooses it; most free prompts still go to Flash-Lite.

For buyers, the practical step is to rerun real workloads before January 1. Teams on Flash face a doubled per-token rate either way. The question is whether Hydrogen's extra skill and lower token use justify it. A cheaper tier or a rival may now fit better.

For investors, the release shows Google pushing customers up its price ladder while undercutting its own premium line on value. It suggests that, at Google at least, mid-range AI is getting better faster than it is getting cheaper.