ModelBrief · Gemini 4 Hydrogen · Topic Page
Last updated Nov 13, 2026

Hydrogen vs Carbon vs Argon: How to Choose, and How Orchestrator Pairs Them

Published Nov 13, 2026 · Last updated Nov 13, 2026

The Gemini 4 family now has three tiers: Carbon, Argon, and Hydrogen. Google's official division of labor: Carbon is the strongest, Argon the balanced one, Hydrogen the fastest. But the AA index made things interesting: Hydrogen at 51 points, one point above Argon's 50. This article synthesizes orcarouter's and the-decoder's comparisons to explain how to actually choose between the three tiers, and how Antigravity's Orchestrator pairs Carbon with Hydrogen.

Three-tier comparison

DimensionGemini 4 HydrogenGemini 4 CarbonGemini 4 Argon
Released2026-11-12 GA2026-11-10 GA2026-09-30
PositioningFast execution tierStrongest reasoning flagshipHard, long-horizon tasks
API price$1.50 / $7.50 (standard)$2 / $10 through Nov 30, then $4 / $20Same as Carbon
AA index51 (high) / 54 (xhigh)60 (high)50 (high)
Default thinking levelhighmediummedium
Cost per task (AA)high $1.87 / xhigh $3.16high $7.76~$1.99 promo / ~$3.98 standard
Free tierYesNoNo
SubscriptionSelectable on Pro / Ultra; auto for free usersUltra onlyPro and above
Best forEveryday coding, batch agents, high-concurrency workersThe hardest tasks: long-horizon agents, tricky debugging, scientific reasoningMost programming and knowledge work, interactive coding

All three have a 1M context window and a 65,536-token sync output cap; Hydrogen is adapted from Carbon.

51 vs 50: basically a tie — don't over-read it

Hydrogen's 51 against Argon's 50 is a 1-point gap — essentially a tie under the AA index's methodology. And the levels aren't symmetric: Hydrogen's 51 is its default high level, while Argon's default is medium — Hydrogen ships at a high level out of the box. On terminal work Hydrogen doesn't lead either: Artificial Analysis measured Terminal-Bench 4.0 at 54%, below Sol's 56% and Sonnet 5.5's 64%. The-decoder's judgment: xhigh's 54 is roughly on par with Sonnet 5.5 and GPT-6.1 Sol (55 each at max), but terminal work is weak.

Hydrogen's real advantages aren't the intelligence score — they're price, the free tier, and the hallucination rate: after December 1 its per-token price is 37.5% of Carbon/Argon's, it is the family's only model with a free tier, and its hallucination rate is just 8% (Sonnet 5.5: 43%, Sol: 52%).

The pricing ladder: December 1 is the dividing line

Today, Hydrogen is only 25% cheaper than Carbon/Argon's $2/$10 — but those are promo prices expiring November 30. After December 1, Carbon/Argon double to $4/$20, and Hydrogen's price is just 37.5% of theirs. Per-task cost (Artificial Analysis): Hydrogen high $1.87, Argon ~$1.99 at promo price and ~$3.98 at standard price — nearly equal during the promo, diverging after December 1. Carbon's high level costs $7.76 per task, more than 4× Hydrogen's high level.

Orchestrator: Carbon directs, Hydrogen works

Antigravity's Orchestrator (launched November 12, 2026 alongside Hydrogen) switches Carbon and Hydrogen dynamically within one session: Carbon first explores dependencies and plans in a read-only phase, Hydrogen writes the edits, updates packages, and runs tests, handing back to Carbon when stuck; in parallel, Carbon coordinates while Hydrogen powers worker and verifier agents in isolated workspaces. Google claims near-Carbon success rates with 24% less quota and 19% faster.

Two caveats, though. First, the benchmark version is self-contradictory — Google's main blog says Terminal-Bench 4.0, the Antigravity post says 4.1, and neither gives an actual success-rate number. Second, access is Ultra subscribers only ("additional models and plans supported in the future"); enterprises get it via Gemini for business. The Antigravity SDK lets developers assign roles directly (demo: Hydrogen for triage and code, Carbon for large refactors, new features, and security checks). The Antigravity post claims "tens of thousands of swarms" have been launched since Carbon arrived, with no exact count.

Decision table

Task typePickWhy
Everyday coding, batch agents, high-concurrency workersHydrogenCheapest, has a free tier, lowest hallucination rate; steadiest in the family in DerpMeter's test
Interactive coding, most knowledge workArgonAA 50 is basically tied with Hydrogen, with faster first tokens
The hardest tasks: long-horizon agents, tricky debugging, scientific reasoningCarbonAA 60, leads all 9 sub-evaluations; $7.76/task is expensive but gets it right first time
Complex projects (planning + execution)Orchestrator (Ultra only) or manual pairingCarbon plans, Hydrogen executes; near-Carbon success at 24% less quota
Tight budget, want to validate firstHydrogen free tierThe family's only free tier; note free-tier data may be used to improve Google's products
Our take: the three tiers aren't a hierarchy — they're a division of labor. Hydrogen's arrival makes "using Carbon for chores" a waste: hand everyday tasks to Hydrogen and save the budget for Carbon on genuinely hard work. The 1-point gap between 51 and 50 isn't worth discussing; what is worth asking is how many of your tasks actually need Carbon.

Reference source: this article is an original rewrite produced by this site; facts and figures were checked against orcarouter.ai: three-tier comparison (Nov 12, 2026)

← Back to the Gemini 4 Hydrogen topic page