Field intelligence for AI-first professionalsVol. II · Nº 195 · Monday, November 16, 2026
← Phantom Notes
Nº 183AI ModelsNovember 12, 20265 min read

Gemini 4 Hydrogen Beats Argon on the Index for Less Money. Its Best Score Needs a Setting That Burns More Tokens Than Carbon, and It Costs Twice What 3.8 Flash Costs Today.

Google launched Gemini 4 Hydrogen on November 12 as the fast member of the Gemini 4 family and the successor to 3.8 Flash: $1.50 per million input tokens and $7.50 output, flat across a 1 million token context, with a free tier in AI Studio. Google's headline is that Hydrogen beats Argon on Artificial Analysis, and AA's pages agree: 51 at the default high setting against Argon's 50, and 54 at xhigh. At xhigh it generated 304 million output tokens on the Index, more than Carbon's 205 million, ranks 78th of 79 for verbosity, and takes 14.8 minutes per task. Per token it costs double Gemini 3.8 Flash's $0.75 and $3.75 through December 31, which is the price 3.8 Flash moves to on January 1. We read Google's post, the AI Studio notes, the pricing page, and AA's model pages.

By T.W. Ghost


Google launched Gemini 4 Hydrogen today, November 12, 2026, the fast and cheap member of the Gemini 4 family and the successor to Gemini 3.8 Flash. The model ID is gemini-4-hydrogen. The paid-tier price is $1.50 per million input tokens and $7.50 per million output, thinking included, flat across the 1 million token window, with implicit caching at $0.075 and no explicit caching. There is a free, rate-limited tier in AI Studio whose prompts Google says are used to improve its products. It is rolling out to the Gemini app, Search, Gemini for business, and Antigravity; free Gemini app users get it only through automatic model selection. Google's headline claim is that Hydrogen beats Gemini 4 Argon on Artificial Analysis's Intelligence Index.

AA's pages say the same thing, and then say what it costs. Hydrogen scores 51 at its default thinking level, high, against Argon's 50 at Argon's maximum; at xhigh it reaches 54. To get the 54 it generated 304 million output tokens on the Index, more than Gemini 4 Carbon at high (205 million) and far more than Argon (113 million), which puts it 78th of 79 on AA's verbosity card, "very verbose" in AA's words, at 14.8 minutes per task against a peer median of 13.1. And the per-token price is exactly double what Gemini 3.8 Flash costs through December 31, $0.75 and $3.75; it matches the rate 3.8 Flash moves to on January 1, 2027. We read Google's launch post, the AI Studio notes, the pricing and deprecations pages, and AA's model pages for Hydrogen, Argon, and Carbon. Here is what held up.


What You Are Paying For

Gemini 4 Hydrogen (November 12, 2026)
Model id; wheregemini-4-hydrogen; Gemini API; AI Studio free tier; Gemini app, Search, Gemini for business, Antigravity
Price, paid tier$1.50 in / $7.50 out per 1M, thinking included; flat across 1M context; implicit cache $0.075; explicit caching not supported
Batch, Flex, ContributorHalf of standard (press, VentureBeat; not rechecked on the pricing tabs)
Free tierFree, rate-limited; data used to improve Google products (paid tier: not used)
3.8 Flash, for comparison$0.75 / $3.75 through Dec 31, 2026; $1.50 / $7.50 from Jan 1, 2027; no shutdown date announced
Thinking levelslow, medium, high (default), xhigh; minimal returns an error; Carbon and Argon default to medium and cap at high
Breaking changes from 3.8 Flashtemperature, top_p, top_k ignored, out-of-range values return 400; no explicit caching; unified 1,048,576-token window; 65,536 output tokens per request unless background mode

The Benchmarks, Sorted by Who Reported Them

ClaimNumberWho
Intelligence Index v5.0.1xhigh 54; high 51; medium 41; low 27Independent (Artificial Analysis)
Argon and Carbon on the same boardArgon 50 at high, $4.97 per task, 113M tokens; Carbon 60 high, 58 medium, 54 low, $5.48 per task at high, 205M tokensIndependent (AA)
Hydrogen at xhigh16th of 80 on the summary card; $3.16 per Index task; 304M output tokens, 78th of 79 for verbosity against a 77.3M median; 14.8 min per task against a 13.1 median; 43.34 s to first answer token; 174 tokens per secondIndependent (AA)
Hydrogen at high8.4 min per task; 110M output tokens; 17.60 s to first answer token; cost per task not shown in the text we readIndependent (AA)
Terminal-Bench 4.1 57% vs 3.8 Flash 18.2%; 2.5x faster than 3.8 Flash at default; more than 5x fewer fabricated claims; HLE-Diamond 58.8%; Vals Tax Agent 52.3% (first); Vals Legal Research 55.8% (tied first); MKA 53.4%; Agents' Last Exam v1.1 71.6%; fifth on arena.ai Agent ArenaVendor (Google)
Antigravity Orchestrator near Carbon on Terminal-Bench 4.0 with 24% less quota and 19% fasterVendor
Safety: overreach down up to 81%; misaligned actions down 12x; 94% internal honesty; 99.9% CoT legibility; 2.3x more resilient on Gray Swan IPIVendor
"Gemini 4 Helix" as a Flash-Lite placeholderUnverified (X leak)

Three honest observations:

  • The 54 is the expensive setting. xhigh means 304 million Index tokens, more than Carbon at high, and slower-than-median tasks. The default, high, is 51.
  • "Beats Argon" is true and narrow. One point, at Hydrogen's default against Argon's maximum. Per task Hydrogen is cheaper, $3.16 at xhigh against Argon's $4.97 at high.
  • Carbon at low already scores 54. The same number as Hydrogen flat out. Compare cost per task, not list price, before picking Hydrogen for quality.

Verified vs Unconfirmed: The Scorecard

ClaimVerdict
Announced Nov 12, 2026; model id; surfacesVerified (Google blog; AI Studio notes)
$1.50 / $7.50; implicit cache $0.075; no explicit caching; free tier trainsVerified (pricing page; AI Studio notes)
Batch, Flex, Contributor at half pricePress (VentureBeat); not rechecked
3.8 Flash $0.75 / $3.75 through Dec 31; $1.50 / $7.50 from Jan 1Verified (pricing and deprecations pages)
Thinking levels and defaults; minimal error; migration breaksVerified (AI Studio notes)
AA scores, tokens, time, cost per taskVerified (AA model pages)
Google's benchmark, speed, and safety figuresVendor
HelixUnverified (X leak only)

What To Do

Benchmark medium and high before defaulting to xhigh; AA shows medium at 41 and 3.8 minutes per task, and the default already costs more thinking tokens than a Carbon or Argon default does.

Skip the upgrade for price until December 31 if you run 3.8 Flash for cost; Hydrogen is priced at 3.8 Flash's 2027 rate with no intro discount, and the quality gap has to pay for twice the per-token bill. Remove cached_content, the sampling parameters, and minimal before switching IDs.

The frontier is elsewhere: Gemini 4 Carbon at 60 is the Google model with the top score, at $2 and $10 through November 30. Our model comparison has the current standings side by side.


Update, November 13, 2026

  • GitHub Copilot. Hydrogen is generally available in GitHub Copilot in the app, the CLI, and VS Code, per GitHub's post.

Sources


*Not sure whether a Flash-class model at next year's Flash price is worth double today? Take the free 2-minute quiz and get matched. Then read about the sibling it edges by one point: Gemini 4 Argon.*

Which model should you be using?

Three minutes, twelve questions, one defensible answer.

Take the quiz →