Decision model

GLiNER2.5-multi-Decide

1 question per request

20.2% of the 2,198 questions right (18.6% to 21.9%, 95% interval). Back to the leaderboard.

How we called it

Kind: Decision model
Access: GLiNER2 server, self-hosted
Paid as: Our own hardware
Model ID: gliner2.5-multi-decide
Parameters: 287M
Hardware: 1 × RTX PRO 6000 Blackwell (96 GB); FP16
Note: Fastino's decision fine-tune of GLiNER2.5 Multi
Confidence from: Option score
Run: 135, 7 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	20.2%	18.6% to 21.9%	444 of 2,198
Reading	21.0%	18.1% to 24.3%	138 of 656
Grammar	19.8%	17.9% to 21.9%	306 of 1,542

Comparison group: H. Compare with another setup.

Accuracy by topic

Topic	Part	Accuracy	Right
Word and phrase meanings	Reading	22.1%	29 of 131
Sentence meanings	Reading	23.4%	36 of 154
How a writer explains and develops ideas	Reading	17.1%	12 of 70
Topic and main point of a paragraph	Reading	22.4%	22 of 98
How a paragraph fits together	Reading	14.2%	15 of 106
Details that support a paragraph’s main point	Reading	24.7%	24 of 97
Word types, such as nouns and verbs	Grammar	17.9%	25 of 140
Groups of words built around a noun	Grammar	12.3%	8 of 65
Verbs and their forms	Grammar	25.3%	23 of 91
Verb forms used like nouns or modifiers	Grammar	12.5%	9 of 72
Word endings and what they do	Grammar	17.7%	14 of 79
How words are built	Grammar	20.3%	14 of 69
The roles of words and phrases in a sentence	Grammar	25.5%	37 of 145
How verbs show who acts and who receives the action	Grammar	24.3%	18 of 74
Types of sentences	Grammar	17.6%	15 of 85
Questions combining several grammar topics	Grammar	17.7%	39 of 220
Sound patterns and changes	Grammar	19.5%	15 of 77
Spelling	Grammar	19.5%	26 of 133
Punctuation	Grammar	23.1%	27 of 117
Problems in sentence wording	Grammar	20.6%	36 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: 23.1 points
AUROC, all questions: 0.54 0.51 to 0.57
AUROC, reading: 0.56 0.51 to 0.61
AUROC, grammar: 0.53 0.49 to 0.57
Brier score: 0.233
Accuracy of its surest answers: most confident fifth: 23.4% right; most confident half: 21.6% right; most confident four fifths: 21.3% right
Typical score of a right answer: 40.7%
Typical score of a wrong answer: 38.5%

Repeats

We did not repeat this setup.

Latency

Latency, median request: 0.009 s (this run)
Latency, 90th percentile: 0.010 s
Requests at once: 1
Throughput: 6615.9 questions per minute of request time
Request time: 0.3 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: 196
Output tokens per question: 0