Chat model

Gemini 4 Hydrogen high · chat

high reasoning · 16 questions per request

Accuracy, all 2,198 questions
98.1%97.4% to 98.6%, 95% interval; 2,156 right
Answers kept at 99% confidence
1,437 of 2,1985 wrong; share right 99.2% to 99.9%, 95% interval
Cost, $ per 1,000 questionsper 1M questions
$4.806$4,806Basis and price under Cost
Latency
31.7 smedian request, 16 questions per request

Back to the leaderboard.

How we called it

Kind
Chat model
Access
Google Antigravity through Can's agbench bridge (Google AI Ultra subscription), one request at a time, thinking high; each request a new Antigravity agent conversation, so Antigravity's own agent instructions and tool declarations (about 13,000 tokens) come before our prompt, with every tool call denied and, as the bridge's owner reports, no workspace files and no earlier conversation; the bridge applies no response format, so the prompt adds a paragraph stating the answer layout
Paid as
Subscription
Model ID
gemini-4-hydrogen
Parameters
not disclosed by makerpages checked: deepmind.google/models/model-cards/gemini-4-hydrogen/; read 12 November 2026
Note
Its English “I don't know” run (1109): one request of 16 questions came back empty twice during the subscription bridge's outage on 14 November 2026; it was asked a third time on 15 November 2026 at the owner's decision, with the same settings, and answered (15 of the 16 right). Both empty replies are kept.
Confidence from
Written estimate
Run
1114, 14 November 2026

Settings

Reasoning
high
Request format
16 questions per request
Requests at once
1
Temperature
not recorded
Run record: harness seed, harness commit and prompt version
Harness seed
0
Harness commit
5424d93c2c6c
Prompt version
254597b8be16

Results

QuestionsAccuracy95% intervalRight
All questions98.1%97.4% to 98.6%2,156 of 2,198
Reading97.3%95.7% to 98.3%638 of 656
Grammar98.4%97.7% to 99.0%1,518 of 1,542

Comparison group: abc. Compare with another setup.

With “Bilmiyorum” (run 1117): Invalid answers: 11; read from broken replies: 5; empty replies asked again: 0. Still invalid: 11 unreadable. Without the rule: 16 invalid. (how)

Accuracy by topic
TopicPartAccuracyRight
Word and phrase meaningsReading97.7%128 of 131
Sentence meaningsReading98.1%151 of 154
How a writer explains and develops ideasReading95.7%67 of 70
Topic and main point of a paragraphReading95.9%94 of 98
How a paragraph fits togetherReading98.1%104 of 106
Details that support a paragraph’s main pointReading96.9%94 of 97
Word types, such as nouns and verbsGrammar99.3%139 of 140
Groups of words built around a nounGrammar98.5%64 of 65
Verbs and their formsGrammar100.0%91 of 91
Verb forms used like nouns or modifiersGrammar98.6%71 of 72
Word endings and what they doGrammar98.7%78 of 79
How words are builtGrammar95.7%66 of 69
The roles of words and phrases in a sentenceGrammar96.6%140 of 145
How verbs show who acts and who receives the actionGrammar98.6%73 of 74
Types of sentencesGrammar98.8%84 of 85
Questions combining several grammar topicsGrammar98.6%217 of 220
Sound patterns and changesGrammar98.7%76 of 77
SpellingGrammar100.0%133 of 133
PunctuationGrammar98.3%115 of 117
Problems in sentence wordingGrammar97.7%171 of 175

English

The Text tab's 547 kept reading questions translated into English, asked with this setup's Text tab settings. Every setup: English reading; against Turkish: English and Turkish.

Right answers
529 of 547, first English run (1107)
Accuracy
96.7%
Non-hallucination with “I don't know”
20.0%: 4 “I don't know” of 20 not right (run 1109)
Answered with “I don't know”
99.3%: 543 of 547
Error rate with “I don't know”
2.9%: 16 of 543 answers
Cost, $ per 1,000 questionsper 1M questions, first English run
4.9274,927subscription, priced as the API
Latency, median request, first English run
28.4 s; 1 request at once
English run
1107
Machine, English runs
hosted API
Date
14 November 2026
Request format
16 questions per request
Reasoning
high
Prompt version
5c3a4849752e
Translations, SHA-256
bf0d47cd385a93050c65906977e199ef7e45d5848dbbdef2bc8d124f4d5abf68

With “I don't know” (run 1109): Invalid answers: 0; read from broken replies: 0; empty replies asked again: 1 (1 came back with answers, 16 answers). Without the rule: 16 invalid. (how)

Confidence

Answers kept at 99% or more
1,437, of which 5 wrong (share right: 99.2% to 99.9%, 95% interval); 761 skipped, the model less than 99% sure
Calibration gap
1.2 points
AUROC, all questions
0.89 0.83 to 0.95
AUROC, reading
0.96 0.93 to 0.99
AUROC, grammar
0.85 0.75 to 0.94
Brier score
0.015
Accuracy of its surest answers
most confident fifth: 99.7% right; most confident half: 99.7% right; most confident four fifths: 99.6% right
Typical score of a right answer
99.0%
Typical score of a wrong answer
86.5%

Repeats

We did not repeat this setup.

Latency

Latency, median request
31.7 s (this run)
Latency, 90th percentile
71.3 s
Requests at once
1
Throughput
24.7 questions per minute of request time
Request time
89.1 minutes, over the requests whose answers were kept

Latency: the time from sending one request to receiving its answer, measured from our computer, network and any retries within the request included. Request time: the time at least one request was in flight, over the requests whose answers were kept. It is not the run's elapsed time: it leaves out requests that failed and any pause before a run was resumed.

Cost

Cost per 1,000 questionsper 1M questions
$4.806$4,806
Whole run
$10.56
Basis
Subscription run priced as if we had used the API
Price
$1.50 input, $0.075 cached input, $7.50 output per 1M tokens (source, 2026-11-12)
Input tokens per question
638
Output tokens per question
513