Google has released Gemini 4 Hydrogen, a new general-purpose AI model positioned as the successor to Gemini 3.8 Flash — and, according to benchmark data, a model that already outperforms the company's own larger Gemini 4 Argon less than a month after that flagship's debut.
The release, announced on Google's official blog under the banner "Introducing Gemini 4 Hydrogen for Agentic Workflows," fills the speed-and-cost tier of Google's lineup while the company's frontier models, Gemini 4 Carbon and Gemini 4 Argon, target deep, long-horizon reasoning on harder problems. Hydrogen is designed for everyday coding, agentic workflows, and high-volume consumer queries — the workloads where latency and price matter more than maximum reasoning depth. For more context on this story, see our ongoing latest AI developments.
Benchmarks: The Smaller Model Wins
The headline surprise is that Hydrogen doesn't just match Argon on quality — it beats it. According to Google's figures reported by The Decoder, Gemini 4 Hydrogen scores 51 on the Artificial Analysis Intelligence Index at its default "high" thinking level, edging out the larger Argon's score of 50. At Hydrogen's newly introduced top reasoning tier, "xhigh," the model reaches 54, trailing only Claude Sonnet 5.5 and GPT-6.1 Sol, which each score 55 at their maximum thinking levels.
That result makes Argon look outdated less than four weeks after its launch. The awkward timing hasn't gone unnoticed: when Google unveiled Carbon just two days before Hydrogen's release, the company was still recommending that developers stick with Argon for daily coding work — advice that Hydrogen immediately rendered moot, as The Decoder observed.
Hydrogen also offers finer-grained control over the speed-quality tradeoff than its predecessor flagship. The model ships with four thinking levels — low, medium, high (the default), and the new xhigh tier — whereas Argon offers only medium and high. For developers running agentic systems at scale, the lower tiers offer a way to cut latency and cost on routine steps while escalating only genuinely hard reasoning to the maximum setting.
Speed and Capability Gains Over 3.8 Flash
Against the model it replaces, the gains are unambiguous. Google says Hydrogen finishes tasks 2.5 times faster than Gemini 3.8 Flash at the default thinking level. On Terminal-Bench 4.1, a benchmark measuring agentic terminal and coding work, the score jumps from 18.2 percent for 3.8 Flash to 57 percent for Hydrogen.
Ars Technica characterized the release as Google "finally replacing" the 3.8 Flash generation, and 9to5Google reported Google claiming "significant leaps" in both speed and intelligence. Crypto Briefing's summary matched the key numbers: 2.5 times faster than 3.8 Flash while edging past Argon on quality indexes.
Pricing and Context Window
Hydrogen's commercial terms put it firmly in the mid-tier price band Google has used to drive volume on its Flash models:
- Input: $1.50 per million tokens
- Output: $7.50 per million tokens
- Cached input: $0.075 per million tokens
The context window is generous on both ends of the pipe: one million tokens of input and — notably — one million tokens of output, a spec that matters for long-running agents that generate extended multi-step traces, codebases, or reports in a single pass.
Where It's Available
The model is live across Google's major surfaces: the Gemini API, the Gemini app, Google Search, and Google Antigravity, the company's agentic development environment. InfoWorld reported that Antigravity has also gained a multi-model Orchestrator feature alongside Hydrogen's arrival, letting developers route tasks across different Gemini models from one interface. GitHub followed within a day, announcing Gemini 4 Hydrogen in GitHub Copilot, which extends the model's reach into the editor where much day-to-day coding happens. French coverage at Generation-NT reported the rollout to all users, replacing 3.8 Flash in the consumer product as well.
Why the Release Matters
The Flash tier is where AI usage volume actually lives. Frontier releases like Gemini 4 Carbon capture headlines, but the workhorse models answer the billions of everyday queries, code completions, and agent steps that make up the real economics of AI serving. By claiming benchmark parity — or better — with its own flagship at Flash-tier prices, Google is applying pressure on rivals whose mid-tier offerings lag their frontier models by a wider gap.
It also sharpens the competitive picture against OpenAI and Anthropic. Hydrogen's xhigh score of 54 sits one point behind Claude Sonnet 5.5 and GPT-6.1 Sol on the Artificial Analysis Intelligence Index, per The Decoder's reporting — close enough that model choice for agentic coding work now hinges as much on price and speed as on raw scores. That dynamic will only intensify as inference costs become the decisive factor in whether agentic products can operate profitably.
For developers evaluating the new model, the practical question is migration: Google's benchmarks suggest 3.8 Flash workloads should move to Hydrogen with better results at lower effective latency, while teams currently using Argon for routine coding may find the smaller, cheaper model is now the stronger choice. You can follow the rest of our Gemini coverage — including Gemini 4 Argon's full release and pricing and its enterprise rollout — for the complete picture of Google's frontier lineup.
---
Stay Ahead of AIGet the latest AI news, analysis, and breakthroughs — all in one place.
Read more AI news →
