Gemini 4 Hydrogen vs Argon vs Carbon: Scores, Price and Which One to Use

2026-11-13 · 5 min read

Geminimodel-release

Gemini 4 Hydrogen is the new default pick for everyday Gemini work, and it makes Argon hard to justify. Google released it on November 12, 2026, as the replacement for Gemini 3.8 Flash. At its default "high" thinking level it scores 51 on the Artificial Analysis Intelligence Index, one point above the larger Gemini 4 Argon (50), and it costs $1.50 per million input tokens and $7.50 per million output tokens. Carbon is still the model for the hardest problems. Hydrogen is not on aiwith.chat yet; Gemini 4 Carbon and Gemini 3.1 Pro are.

Published November 13, 2026. Numbers come from Google's announcement as reported by The Decoder and from Artificial Analysis. This is not a hands-on review.

What Google released

Hydrogen is Google's fast, cheap general-purpose model, the slot the Flash series used to fill. The main specs:

Thinking levels: four (low, medium, high, xhigh). High is the default. Argon only has medium and high.
Context: 1 million input tokens, and it can also write up to 1 million output tokens.
Price: $1.50 per million input tokens, $7.50 per million output tokens, $0.075 per million for cache hits.
Availability: Gemini API (AI Studio, Android Studio), the Gemini app, Google Search, Gemini for Business, and Google Antigravity for Pro and Ultra subscribers.

Against the previous Flash generation the jump is large. Google says Hydrogen finishes tasks 2.5 times faster than Gemini 3.8 Flash, scores 57% on Terminal-Bench 4.1 where Flash scored 18.2%, and hallucinates five times less.

Google also made Hydrogen the model behind Google Search, calling it "the biggest model upgrade to Search yet". Search now handles multi-part questions and typos better, fans out to more sources, and shows more tables and images in answers.

Hydrogen vs Argon vs Carbon

Google's Gemini 4 lineup now has three models. Here is how they line up on the numbers that matter:

	Gemini 4 Hydrogen	Gemini 4 Argon	Gemini 4 Carbon
Released	Nov 12, 2026	Oct 14, 2026	Oct 28 (GA Nov 10)
AA Intelligence Index	51 (high), 54 (xhigh)	50	60
Thinking levels	low / medium / high / xhigh	medium / high	Not published
Price per 1M tokens (in / out)	$1.50 / $7.50	Same list price as Carbon	$2 / $10 intro, $4 / $20 standard
Cost per AA task	$1.87 (high), $3.16 (xhigh)	Not published	$7.76
Built for	Everyday coding, agents, search	General work	Hardest multi-step coding and research

Two things stand out.

Argon has lost its reason to exist. When Carbon launched two days earlier, Google still recommended Argon for daily coding. Hydrogen now scores higher than Argon and costs less, less than four weeks after Argon came out. If you set up a workflow on Argon in October, move it to Hydrogen.

Carbon is still a tier above. Carbon's 60 on the index is far ahead of Hydrogen's 54 at its best setting. Carbon also leads coding benchmarks like CursorBench 4.0 and finds real security vulnerabilities, which Hydrogen is not pitched for. Google's own Antigravity coding tool now switches between the two inside one session: Carbon for hard steps, Hydrogen for the rest.

How it stacks up outside Google

Artificial Analysis puts Hydrogen at xhigh roughly level with Claude Sonnet 5.5. At the default high setting (51) it sits just behind GPT-6.1 Sol and Claude Opus 5.5 (both 52).

Hallucinations: only 8% on AA-Omniscience, which is a real strength for research and summaries.
Terminal and agent coding: weaker. Hydrogen scores 54% on Terminal-Bench 4.0 and 50% on Terminal-Bench-Science, against 64% and 53% for Claude Sonnet 5.5 (max) and 56% and 58% for GPT-6.1 Sol (max).
Cost: about $3.16 per task at xhigh, roughly a third of Claude Sonnet 5.5 (max) at similar intelligence, but about 2.5 times what GPT-6.1 Sol (max) costs.

Watch the thinking level. Medium drops to 41 on the index and low to 27. xhigh uses nearly twice the output tokens of high, so most people should stay on the default.

Is it worth switching?

You use Argon today: switch to Hydrogen. It is better and cheaper.
You use Gemini 3.x Flash for cheap, fast tasks: switch. The quality gap is large and the price is still low.
You need the strongest Gemini for hard coding or research: stay on Carbon.
You run coding agents in a terminal all day: Claude Sonnet 5.5 or GPT-6.1 Sol still do better on those benchmarks.
You mostly write, summarize and research: Hydrogen's low hallucination rate and 1M-token output make it a good fit once you can get it.

Try Gemini on aiwith.chat

aiwith.chat does not offer Gemini 4 Hydrogen yet. We are working on adding it, and this post will be updated when it is live.

You can already use Gemini 4 Carbon and Gemini 3.1 Pro on aiwith.chat in the same chat window, with no Google account needed. Carbon is the stronger model of the two and the right choice for hard problems; Gemini 3.1 Pro is fine for everyday questions.

Chat with Gemini on aiwith.chat