Chat model

GLiNER2.5 Multi (base)

1 question per request

17.5% of the 2,198 questions right (15.9% to 19.1%, 95% interval). Back to the leaderboard.

How we called it

Kind: Chat model
Access: GLiNER2 server, self-hosted
Paid as: Our own hardware
Model ID: gliner2.5-multi-v1
Parameters: 287M
Hardware: 1 × RTX PRO 6000 Blackwell (96 GB); FP16
Note: Multilingual extraction and classification model
Confidence from: Option score
Run: 122, 5 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	17.5%	15.9% to 19.1%	384 of 2,198
Reading	17.2%	14.5% to 20.3%	113 of 656
Grammar	17.6%	15.8% to 19.6%	271 of 1,542

Comparison group: H. Compare with another setup.

Accuracy by topic

Topic	Part	Accuracy	Right
Word and phrase meanings	Reading	18.3%	24 of 131
Sentence meanings	Reading	13.0%	20 of 154
How a writer explains and develops ideas	Reading	17.1%	12 of 70
Topic and main point of a paragraph	Reading	21.4%	21 of 98
How a paragraph fits together	Reading	20.8%	22 of 106
Details that support a paragraph’s main point	Reading	14.4%	14 of 97
Word types, such as nouns and verbs	Grammar	16.4%	23 of 140
Groups of words built around a noun	Grammar	13.8%	9 of 65
Verbs and their forms	Grammar	17.6%	16 of 91
Verb forms used like nouns or modifiers	Grammar	15.3%	11 of 72
Word endings and what they do	Grammar	21.5%	17 of 79
How words are built	Grammar	13.0%	9 of 69
The roles of words and phrases in a sentence	Grammar	18.6%	27 of 145
How verbs show who acts and who receives the action	Grammar	20.3%	15 of 74
Types of sentences	Grammar	22.4%	19 of 85
Questions combining several grammar topics	Grammar	14.5%	32 of 220
Sound patterns and changes	Grammar	16.9%	13 of 77
Spelling	Grammar	13.5%	18 of 133
Punctuation	Grammar	14.5%	17 of 117
Problems in sentence wording	Grammar	25.7%	45 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: 35.6 points
AUROC, all questions: 0.45 0.42 to 0.48
AUROC, reading: 0.42 0.36 to 0.48
AUROC, grammar: 0.46 0.42 to 0.50
Brier score: 0.298
Accuracy of its surest answers: most confident fifth: 15.5% right; most confident half: 15.0% right; most confident four fifths: 16.5% right
Typical score of a right answer: 49.7%
Typical score of a wrong answer: 53.1%

Repeats

We did not repeat this setup.

Latency

Latency, median request: 0.032 s (this run)
Latency, 90th percentile: 0.036 s
Requests at once: 1
Throughput: 1922.6 questions per minute of request time
Request time: 1.1 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

$ per 1,000 questions: Not measured
Basis: No dollar estimate; see hardware and time
Input tokens per question: 196
Output tokens per question: 0