Chat model

Qwen3-14B

reasoning off · 16 questions per request

45.7% of the 2,198 questions right (43.6% to 47.8%, 95% interval). Back to the leaderboard.

How we called it

Kind: Chat model
Access: vLLM, self-hosted, BF16
Paid as: Our own hardware
Model ID: qwen3-14b
Parameters: 14.8B
Hardware: 1 × RTX PRO 6000 Blackwell (96 GB); BF16
Note: Erk-14B's base model
Confidence from: Answer letter probability
Run: 131, 6 October 2026

Settings

Reasoning: off
Request format: 16 questions 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	45.7%	43.6% to 47.8%	1,005 of 2,198
Reading	72.1%	68.6% to 75.4%	473 of 656
Grammar	34.5%	32.2% to 36.9%	532 of 1,542

Comparison group: E. Compare with another setup.

Accuracy by topic

Topic	Part	Accuracy	Right
Word and phrase meanings	Reading	61.1%	80 of 131
Sentence meanings	Reading	74.7%	115 of 154
How a writer explains and develops ideas	Reading	65.7%	46 of 70
Topic and main point of a paragraph	Reading	85.7%	84 of 98
How a paragraph fits together	Reading	67.9%	72 of 106
Details that support a paragraph’s main point	Reading	78.4%	76 of 97
Word types, such as nouns and verbs	Grammar	32.1%	45 of 140
Groups of words built around a noun	Grammar	29.2%	19 of 65
Verbs and their forms	Grammar	29.7%	27 of 91
Verb forms used like nouns or modifiers	Grammar	33.3%	24 of 72
Word endings and what they do	Grammar	30.4%	24 of 79
How words are built	Grammar	36.2%	25 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	33.8%	25 of 74
Types of sentences	Grammar	31.8%	27 of 85
Questions combining several grammar topics	Grammar	27.7%	61 of 220
Sound patterns and changes	Grammar	31.2%	24 of 77
Spelling	Grammar	37.6%	50 of 133
Punctuation	Grammar	45.3%	53 of 117
Problems in sentence wording	Grammar	52.0%	91 of 175

Confidence

Answers kept at 99% or more: 975, of which 342 wrong (share right: 61.9% to 67.9%, 95% interval); 1,223 skipped, the model less than 99% sure
Calibration gap: 41.0 points
AUROC, all questions: 0.73 0.71 to 0.75
AUROC, reading: 0.81 0.77 to 0.84
AUROC, grammar: 0.62 0.59 to 0.65
Brier score: 0.400
Accuracy of its surest answers: most confident fifth: 86.8% right; most confident half: 62.3% right; most confident four fifths: 50.7% right
Typical score of a right answer: 100.0%
Typical score of a wrong answer: 90.0%

Repeats

We did not repeat this setup.

Latency

Latency, median request: 2.85 s (this run)
Latency, 90th percentile: 3.11 s
Requests at once: 1
Throughput: 331.1 questions per minute of request time
Request time: 6.6 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: 253
Output tokens per question: 8