Decision model

Laya

1 question per request

19.0% of the 2,198 questions right (17.4% to 20.7%, 95% interval). Back to the leaderboard.

How we called it

Kind: Decision model
Access: Laya server, self-hosted
Paid as: Our own hardware
Model ID: laya
Parameters: 421M
Hardware: 1 × RTX PRO 6000 Blackwell (96 GB)
Note: Encoder with a decision head
Confidence from: Option score
Run: 51, 4 October 2026

Settings

Reasoning: off
Request format: 1 question per request
Requests at once: 1
Temperature: not recorded
Harness seed: 0
Harness commit: not recorded
Prompt version: not recorded

Results

Questions	Accuracy	95% interval	Right
All questions	19.0%	17.4% to 20.7%	418 of 2,198
Reading	20.9%	17.9% to 24.2%	137 of 656
Grammar	18.2%	16.4% to 20.2%	281 of 1,542

Comparison group: H. Compare with another setup.

Accuracy by topic

Topic	Part	Accuracy	Right
Word and phrase meanings	Reading	25.2%	33 of 131
Sentence meanings	Reading	22.7%	35 of 154
How a writer explains and develops ideas	Reading	11.4%	8 of 70
Topic and main point of a paragraph	Reading	18.4%	18 of 98
How a paragraph fits together	Reading	26.4%	28 of 106
Details that support a paragraph’s main point	Reading	15.5%	15 of 97
Word types, such as nouns and verbs	Grammar	12.9%	18 of 140
Groups of words built around a noun	Grammar	15.4%	10 of 65
Verbs and their forms	Grammar	16.5%	15 of 91
Verb forms used like nouns or modifiers	Grammar	13.9%	10 of 72
Word endings and what they do	Grammar	16.5%	13 of 79
How words are built	Grammar	14.5%	10 of 69
The roles of words and phrases in a sentence	Grammar	22.8%	33 of 145
How verbs show who acts and who receives the action	Grammar	16.2%	12 of 74
Types of sentences	Grammar	27.1%	23 of 85
Questions combining several grammar topics	Grammar	21.4%	47 of 220
Sound patterns and changes	Grammar	22.1%	17 of 77
Spelling	Grammar	16.5%	22 of 133
Punctuation	Grammar	14.5%	17 of 117
Problems in sentence wording	Grammar	19.4%	34 of 175

Confidence

Answers kept at 99% or more: 0, of which 0 wrong; 2,198 skipped, the model less than 99% sure
Calibration gap: 13.2 points
AUROC, all questions: 0.48 0.45 to 0.51
AUROC, reading: 0.49 0.43 to 0.54
AUROC, grammar: 0.48 0.44 to 0.52
Brier score: 0.186
Accuracy of its surest answers: most confident fifth: 14.4% right; most confident half: 18.0% right; most confident four fifths: 18.8% right
Typical score of a right answer: 28.6%
Typical score of a wrong answer: 29.1%

Repeats

Run 5 times on all questions.

Answers changed in 5 runs: 0 of 2,198
Same wrong answer in all 5: 1,780 of 2,198
Wrong in at least one run: 1,780 of 2,198
Accuracy in each run: 19.0%, 19.0%, 19.0%, 19.0%, 19.0%

Latency

Latency, median request: 0.011 s (its repeated runs)
Latency, 90th percentile: 0.012 s
Requests at once: 1
Throughput: 4612.5 questions per minute of request time
Request time: 0.5 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

$ per 1,000 questions: Not measured
Basis: No dollar estimate; see hardware and time
Input tokens per question: 403
Output tokens per question: 0