RELEASE REPORT · NOVEMBER 12, 2026

Gemini 4 Hydrogen: release report

Google released Gemini 4 Hydrogen on November 12, 2026. Specs, price, where it ranks, how it plays chess, how it solves sudoku, how it plays Mines, how it solves nonograms, how it plays FreeCell, how it plays Starjar, how it plays Nations Royale, how it commands a Wildhaven raid, how it plays Hustle Sim, how it plays Elemental Armory, how it drives Micro Mouse and how it plays Ravelin, with more added as evidence arrives.

DATA AS OFNov 20, 2026
SOURCELab announcement
PUBLISHEDNov 12, 2026

The release

Google released Gemini 4 Hydrogen on November 12, 2026. Read the announcement.

  • Takes unknown. Returns unknown.
  • Not a reasoning model.
  • Closed weights.

Google's release before this was Gemini Decision 1 on November 11, 2026, 1 days earlier. Over the past year it has shipped a model every 6 days on median (41 releases).

Price

$1.50 per million input tokens and $7.50 per million output tokens on Google's API. Cached input is $0.075.

Where it ranks

Added November 20, 2026.

Epoch AI's Capabilities Index puts it at 166.8 (95% interval 164.0 to 170.1), number 9 of 258 models Epoch has scored. The frontier reached this level 2.4 months before it shipped.

See it on Rankings.

Our chess benchmark

Added November 19, 2026.

Rated 1,544 (95% interval 1,341 to 1,747) over 19 games against calibrated Stockfish, number 2 of 18 models we've tested. Standard chess 1,736, Chess960 1,311.

In standard chess: 91.2 accuracy in open positions, 1.7 blunders per 100 moves, 1.0% of answers were illegal moves.

Each game took 12.3 minutes of model time. At the lab's list price its tokens would cost $0.99 a game.

Every game, move by move.

Our sudoku benchmark

Added November 19, 2026.

Scored 95 out of 100 (95% interval 86 to 100), number 3 of 12 models we've tested. It solved 40 of 42 grids and filled 95.2% of blank cells correctly.

By tier, from singles to grids no human technique finishes: Cozy 6 of 6, Gentle 6 of 6, Steady 5 of 6, Tricky 6 of 6, Devious 5 of 6, Expert 6 of 6, Beyond 6 of 6.

It said it was 95% sure on average.

Each grid took 102 seconds and 24,155 output tokens on average. At the lab's list price its tokens would cost $0.19 a grid.

Every answer, cell by cell.

Our Mines benchmark

Added November 19, 2026.

Scored 71 out of 100 (95% interval 56 to 85), number 2 of 9 models we've tested. It won 17 of 24 no-guess boards and cleared 81% of the safe cells on average.

By size: Beginner 6 of 6, Intermediate 6 of 6, Expert 3 of 6, Master 2 of 6.

It left 4 flags on safe cells; 0.1% of its actions were refused as illegal.

Each board took 299 seconds of model time on average. At the lab's list price its tokens would cost $0.591 a board.

Every board, turn by turn.

Our Nonograms benchmark

Added November 19, 2026.

Scored 47 out of 100 (95% interval 36 to 58), number 5 of 9 models we've tested. It solved 17 of 36 puzzles, got 71.9% of cells right and satisfied 60.5% of row and column clues.

By tier: 5 by 5 6 of 6, 10 by 10 6 of 6, 15 by 15 4 of 6, 20 by 20 1 of 6, 25 by 25 0 of 6, Beyond 0 of 6.

It said it was 66% sure on average; 7 wrong answers came with 90% confidence or more.

Each puzzle took 291 seconds of model time on average. At the lab's list price its tokens would cost $0.540 a puzzle.

Every answer, cell by cell.

Our FreeCell benchmark

Added November 19, 2026.

Scored 30 out of 100 (95% interval 17 to 43), number 4 of 9 models we've tested. It won 9 of 30 deals and got 20 of 52 cards home on average.

By tier: Easy 3 of 6, Four cells 4 of 6, Three cells 2 of 6, Two cells 0 of 6, Beyond 0 of 6.

Planning the whole game in one reply, it won 2 of 5.

The referee refused 1.1 of its moves per game.

Each game took 276 seconds of model time on average. At the lab's list price its tokens would cost $0.472 a game.

Every game, move by move.

Our Starjar benchmark

Added November 19, 2026.

Scored 103 against the reference bot's 100 (95% interval 96 to 110), number 5 of 10 models we've tested. It made 1,214 points a game, beat the bot on 4 of 6 jars and never overflowed the jar.

By tier: Daily opening 95, Fresh jar 102, Midgame 108.

Each drop took 16 seconds of model time on average. At the lab's list price its tokens would cost $2.84 a game.

Every jar, drop by drop.

Our Nations Royale benchmark

Added November 19, 2026.

Scored 92 out of 100 (95% interval 73 to 100), number 1 of 10 models we've tested. It won 11 of 12 wars against the game's 47 commanders and held 72% of the world at the end on average. Of the rest, none were still undecided at 15 minutes and 1 was lost.

By tier: Easy 3 of 3, Medium 3 of 3, Hard 3 of 3, Extreme 2 of 3.

War by war it scored 42 points above the game's reference player and 60 points above the same player held to its own clock.

The game refused 0.1 of its orders per decision on average.

Each decision took 18 seconds of model time on average. At the lab's list price its tokens would cost $1.43 a war.

Every war, decision by decision.

Our Wildhaven benchmark

Added November 19, 2026.

Scored 83% destruction (95% interval 77 to 88), number 4 of 9 models we've tested. It earned 2.3 stars a raid, took all three stars in 19 of 48 raids and brought the Townheart down in 46.

By tier: Hunt 96, Reckoning 80, Outmatched 72.

Raid by raid it destroyed 25 points above the game's own raiders and 4 points below a search that tries 64 plans in the simulator.

Each request took 25 seconds of model time on average. At the lab's list price its tokens would cost $0.106 a raid.

Every raid, look by look.

Our Hustle Sim benchmark

Added November 19, 2026.

Scored 75 against the game's best line's 100 (95% interval 47 to 102), number 2 of 13 models we've tested. It reached the goal before its replies ran out in 6 of 6 runs, 1.4 times the best line's game time on average.

By tier: Fresh start 82, Takeover 74, Empire 68.

Each decision took 22 seconds of model time on average. At the lab's list price its tokens would cost $0.757 a run.

Every run, decision by decision.

Our Elemental Armory benchmark

Added November 19, 2026.

Scored 48 out of 100 (95% interval 40 to 56), number 8 of 9 models we've tested. It won 72 of 150 Bounce Duels matches against the game's rival and 49% of the rounds.

By level: Level 1 28 of 50, Level 5 19 of 50, Level 8 25 of 50.

For scale, the rival playing its seat scores 49 and a player who measured every card scores 68.

Its cards averaged 3.0 stars out of 3.

Each decision took 14 seconds of model time on average. At the lab's list price its tokens would cost $0.079 a match.

Every match, round by round.

Our Micro Mouse benchmark

Added November 19, 2026.

Scored 82 against the median person at 100 (95% interval 62 to 102), number 1 of 9 models we've tested. It was faster than the median person on 0 of the 6 daily mazes people raced, and the game's bot scores 94 on them.

On the campaign mazes it scored 87 against the game's bot at 100, and it reached the dock in 14 of its 14 runs.

By tier: Daily mazes 82, Workbench 92, Cold Storage 77, Fast Lane 87, Night Shift 92.

The game's test pilot, which steers every step along a route it is given, scores 48 on the daily mazes.

Each reply took 16 seconds of model time on average. At the lab's list price its tokens would cost $0.646 a run.

Every run, reply by reply.

Our Ravelin benchmark

Added November 19, 2026.

Scored 99 against par at 100 (95% interval 62 to 136), number 1 of 10 models we've tested. It held 5 of 6 boards and beat par on 5, losing 1.7 lives a battle; the game's own bot scores 76 on the same boards.

By tier: Daily siege 120, Campaign 119, Iron 57.

Each decision took 17 seconds of model time on average. At the lab's list price its tokens would cost $0.327 a battle.

Every battle, decision by decision.

Each section was written on the date shown and is not revised afterwards. Specifications from Models.dev, prices from provider listings, capability scores from Epoch AI under CC BY 4.0, chess, sudoku and Crest Arcade game results from Glimmer Index Bench.