Skip to the yard
Waxsealai

Google Gemini 4 Hydrogen Costs Less Than Argon and Scores Higher on Key Benchmarks

Shipped
Shipper
Duncan Riley
Readers
379

Gemini 4 Hydrogen posts higher reasoning and coding scores than Argon at lower token prices, with a new high thinking setting and broad access.

Google Gemini 4 Hydrogen Costs Less Than Argon and Scores Higher on Key Benchmarks

Gemini 4 Hydrogen gives developers a lower priced option in Google’s latest model family, and Google reports that it edges past the more expensive Gemini 4 Argon on one intelligence index. The reported gap is narrow at the default setting: Hydrogen scores 51 to Argon’s 50 on the Artificial Analysis Intelligence Index. A new “xhigh” thinking level raises Hydrogen’s score to 54.

Those results make Hydrogen’s practical case less about a clear win on every measure and more about the mix of price, speed and access. Google describes the model as “built for rapid execution,” in a statement from Google DeepMind head Koray Kavukcuoglu. The benchmarks cited by Google show a larger separation from Gemini 3.8 Flash, the September model used as the comparison point for most of the figures.

Speed and command line performance

At its default setting, Hydrogen completed tasks 2.5 times faster than Gemini 3.8 Flash, according to Google. On Terminal-Bench 4.1, a test of command line work, Hydrogen scored 57%. Gemini 3.8 Flash scored 18.2% on that benchmark, so Hydrogen more than tripled the older model’s result.

These figures point to a model aimed at tasks where response time and tool use matter alongside answer quality. They don’t show how Hydrogen performs across every coding workload, or how much of the measured advantage holds under different prompts, task mixes and compute settings. Developers evaluating it will need to compare results on their own workflows, especially where small benchmark differences may not translate into faster or more reliable work.

Google is also bringing Hydrogen into its Antigravity coding tool through Orchestrator. The feature chooses between Hydrogen and Gemini 4 Carbon as a session progresses. In Google’s runs of Terminal-Bench 4.0, Orchestrator came close to Carbon’s success rate while using an average of 24% less quota. That could help teams reserve a higher priced model for tasks that benefit from it, while assigning other steps to Hydrogen. The reported average does not establish how much quota an individual project would save.

Broader access and lower token prices

Hydrogen is available through the Gemini API, Google AI Studio and Google Search. People using the free Gemini app can access it as well. Paid Google AI Pro and Ultra subscribers can use it in Antigravity.

Across its full 1 million-token context window, Hydrogen costs $1.50 per million input tokens and $7.50 per million output tokens. Argon’s current introductory rates are $2 per million input tokens and $10 per million output tokens. Those Argon prices are scheduled to end Nov. 30, after which they rise to $4 and $20, respectively. For workloads that produce many output tokens, that difference can shape the cost of repeated model calls as much as the choice of model itself.

The rollout also differs from Argon’s. Argon spent nearly two weeks with members of Google’s Fairwind Program, a group of vetted cybersecurity defenders, before reaching paying developers on Oct. 13. Hydrogen skipped that early access stage and is available to a broader set of users from launch. Carbon, which opened to paying developers and Google AI Ultra subscribers two days before Hydrogen, also began with Fairwind members.

What the early evidence shows

Google’s early testers include Andrew Carr, a former OpenAI scientist and current chief scientist at 3D animation startup Cartwheel. Carr said Hydrogen identified two problems in an approach his team had been stuck on for 10 days. That account offers a concrete example of the model helping with a difficult workflow, though one reported experience cannot establish how often users will see similar gains. Discovery Loop, a startup co-founded by former Google Chief Scientist Jeff Dean after he left in August, was another early tester.

Google has released three Gemini 4 models in just over six weeks, following Kavukcuoglu’s late September comment that the company wanted the family out “much earlier” than year-end. Hydrogen arrived two days after Carbon became available to paying developers and Ultra subscribers. The pace gives developers several models to assess within a short period, while making clear comparisons dependent on matching each model’s settings, task types and prices.

For teams considering adoption, Hydrogen’s clearest advantages in Google’s reported results are its lower cost than Argon, broader availability and strong scores against Gemini 3.8 Flash on speed and command line coding. The index lead over Argon is only one point at the default setting, rising to four points with xhigh. The next useful evidence will be how those results hold up on independent tests and production workloads, where latency, output costs and task success matter together.

Source: Duncan Riley · siliconangle.com

Same vesselRelated cargo