Gemini 4 Hydrogen released: LiveBench 77.0, League #13 at $1.50/$7.50 per million tokens.
Gemini 4 Hydrogen was released on November 12, 2026, and enters the AI Pro League at #13 with a League Score of 85.0 — built from a LiveBench 77.0 — at $1.50/$7.50 per million tokens, exactly 25% below Gemini 4 Carbon and Gemini 4 Argon on every rate. This post works the normalisation maths in full, prices the family on standard and cached-heavy workloads, and states what a single-component record does and does not prove.
By AI Pro League · Published

Gemini 4 Hydrogen has a League Score, and it lands mid-table rather than at the frontier. Our November 13, 2026 changelog records the release from Google’s official announcement (blog.google, read today) and the Gemini API pricing page (ai.google.dev): released Nov 12, 2026, adapted from Gemini 4 Carbon, priced at $1.50 input / $7.50 output per million tokens with cached input at $0.075, via Google AI Studio, Gemini API. Its first tracked score is 77.0 on LiveBench 2.0 (results dated 2026-11-06; read today) — #12 on that board — which normalises to a League Score of 85.0 and overall #13 of 51 scored models. What follows is computed from the database at build time.
Gemini 4 Hydrogen: the record, dated and priced
| Measure | Gemini 4 Hydrogen |
|---|---|
| Release date | Nov 12, 2026 |
| LiveBench raw score (LiveBench 2.0) | 77.0 (#12 of 34) |
| League Score / overall rank | 85.0 / #13 of 51 scored models (64 tracked) |
| Value rank | #14 |
| Score coverage | 40% — one component (LiveBench) |
| Input / output per 1M tokens | $1.50 / $7.50 |
| Cached input per 1M tokens | $0.075 |
| Blended price per 1M tokens | $3.00 |
| Context window / max output | 1,048,576 / 1,048,576 tokens |
| Inputs | Text, Image, Audio, Video |
| Knowledge cutoff | March 2026 |
Two record details deserve plain words. The max output in the database — the full 1,048,576-token window — is the background long-decode figure from the November 13 changelog entry; the default synchronous output is 64K. And Google’s announcement leads with an AA Index of 51 at high and 54 at xhigh: that index is not a League component, and this site does not score it. The tracked score behind the #13 ranking is LiveBench, and LiveBench only — the same single-component shape as its sibling’s debut, priced in our Gemini 4 Carbon first-score analysis.
How 77.0 becomes a League Score of 85.0
Hydrogen’s League Score rests on this single component, so it is checkable by hand. Normalisation stretches the recorded LiveBench field — 34 models, raw scores from 32.4 to 84.9, median 68.9 — across a 0–100 scale: (77.0 − 32.4) ÷ 52.5 × 100 = 85.0. With one recorded component at 40% weight, the League Score is that normalised value: 85.0. LiveBench carries 40% of the League Score — the joint-largest weight of the five components — for the reasons set out in our LiveBench explainer; the full field sits on the LiveBench leaderboard.
| Rank | Model | League Score | Coverage | Workload cost |
|---|---|---|---|---|
| 11 | Gemini 4 Argon | 86.3 | 40% | $40.00 |
| 12 | Claude Sonnet 5.5 | 85.0 | 40% | $40.00 |
| 13 | Gemini 4 Hydrogen (new) | 85.0 | 40% | $30.00 |
| 14 | DeepSeek V4.1 Flash | 84.4 | 55% | $5.40 |
| 15 | Claude Sonnet 5 | 78.9 | 75% | $40.00 |
Hydrogen ties the model one place above it on score and sits 85.0 points below its sibling Gemini 4 Argon — itself a single-component record — while the standings as a whole live in the AI model rankings.
The Gemini family, priced: exactly 25% below its siblings
| Model | League Score | $/1M in / out / cached | Workload cost |
|---|---|---|---|
| Gemini 4 Hydrogen | 85.0 (#13) | $1.50 / $7.50 / $0.075 | $30.00 |
| Gemini 4 Carbon | 97.3 (#3) | $2.00 / $10.00 / $0.10 | $40.00 |
| Gemini 4 Argon | 86.3 (#11) | $2.00 / $10.00 / $0.10 | $40.00 |
| Gemini 3.8 Flash | 60.6 (#33) | $0.75 / $3.75 / $0.075 | $15.00 |
The family relationship is exact, not approximate: Hydrogen’s input, output and cached rates are each 25% below Carbon’s and Argon’s — $1.50 against $2.00, $7.50 against $10.00, $0.075 against $0.10 — so the standard workload falls by the same quarter, from $40.00 to $30.00. Against its predecessor the move runs the other way: Hydrogen’s headline rates are exactly twice Gemini 3.8 Flash’s recorded $0.75 / $3.75, and its workload is twice Flash’s $15.00 — the upgrade path costs double the model it upgrades from, at 24.4 more League points.
What the December 1 step and the cache do to the bill
Two dated facts change how the table above reads. First, Carbon and Argon step to $4.00 / $20.00 on December 1, 2026, per the November 11, 2026 changelog (Google’s official pricing page) — their standard workload becomes $80.00, so Hydrogen’s $30.00 workload goes from a quarter below its siblings to well under half, on unchanged token counts and unchanged recorded Hydrogen rates. Second, the cache: at $0.075 per million, cached tokens cost 95% less than fresh input, so a cached-heavy run of 8M cached + 2M fresh input + 2M output costs $18.60 on Hydrogen against $24.80 on Carbon at today’s rates — the pattern priced across the field in our prompt caching pricing analysis. For scale, the League leader, Claude Fable 5.2, runs the standard workload at $160.00 — more than five times Hydrogen’s bill. The full rate card sits in our AI model pricing index.
Value rank #14 against overall rank #13
Hydrogen’s two ranks nearly coincide: #13 overall, #14 on value — League Score per blended dollar, at a blended $3.00 per million tokens. That near-identity is itself the finding. The frontier pattern runs the other way: Claude Fable 5.2 is #1 overall and #33 on value, a 32-place collapse, because a perfect score at a frontier price buys few points per dollar. Hydrogen gives up 15.0 points to the leader and keeps almost all of its standing when price enters the maths — a mid-table score at a mid-table price behaves like one on both tables.
What the #13 ranking does not yet prove
- The record is one benchmark deep. Coverage is 40% — LiveBench only. No GPQA Diamond, SWE-bench Verified, MMLU-Pro or LMArena result is recorded yet, so 85.0 is a finding about one benchmark, not about the other four — the standing caveat in our score coverage explainer. A promising rumour vs a finding: treat the rank as the rumour until a second component lands.
- Its predecessor’s record is deeper and scores lower. Gemini 3.8 Flash carries 55% coverage — LiveBench plus a GPQA Diamond component — and scores 60.6. Hydrogen’s higher number rests on the thinner file; if its second component normalises below its LiveBench component, the composite falls, exactly as a deeper record pulled other single-benchmark debuts down before it.
- The headline number is not ours. Google’s AA Index — 51 at high, 54 at xhigh — is the figure the announcement leads with, and it plays no part in the League Score. A reader comparing 51 or 54 against the 85.0 on this page is comparing two different instruments.
- The family gap is about to widen on paper. The December 1 step doubles Carbon’s and Argon’s rates while Hydrogen’s recorded rates stand — relative value improves without Hydrogen changing at all, and any budget comparison written this week should say which side of December 1 it prices.
None of that dims the release itself: a model adapted from the #3 frontier entry, at exactly a quarter below it on every rate, entering at #13 on its first tracked score is a clean value proposition — on one benchmark. The score to watch is the second one: whichever tracked leaderboard reports on Hydrogen next will test whether 85.0 is a ceiling or a starting point. The model’s full record lives on its Gemini 4 Hydrogen page, and the standings themselves live in the AI model rankings.