Chat model

GLM-5.3-Flash

reasoning off · 16 questions per request

57.8% of the 2,198 questions right (55.7% to 59.8%, 95% interval). Back to the leaderboard.

How we called it

Kind: Chat model
Access: TensorFold, self-hosted, EXL3 4-bit experts
Paid as: Our own hardware
Model ID: glm-5.3-flash
Parameters: 320B (18B active per token)
Hardware: 4 × RTX PRO 6000 Blackwell (96 GB); EXL3 4-bit experts, FP8 dense layers and KV cache
Confidence from: Written estimate
Run: 159, 11 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	57.8%	55.7% to 59.8%	1,270 of 2,198
Reading	82.2%	79.1% to 84.9%	539 of 656
Grammar	47.4%	44.9% to 49.9%	731 of 1,542

Comparison group: wxyzAB. Compare with another setup.

Accuracy by topic

Topic	Part	Accuracy	Right
Word and phrase meanings	Reading	76.3%	100 of 131
Sentence meanings	Reading	84.4%	130 of 154
How a writer explains and develops ideas	Reading	62.9%	44 of 70
Topic and main point of a paragraph	Reading	92.9%	91 of 98
How a paragraph fits together	Reading	78.3%	83 of 106
Details that support a paragraph’s main point	Reading	93.8%	91 of 97
Word types, such as nouns and verbs	Grammar	52.1%	73 of 140
Groups of words built around a noun	Grammar	41.5%	27 of 65
Verbs and their forms	Grammar	53.8%	49 of 91
Verb forms used like nouns or modifiers	Grammar	48.6%	35 of 72
Word endings and what they do	Grammar	48.1%	38 of 79
How words are built	Grammar	30.4%	21 of 69
The roles of words and phrases in a sentence	Grammar	38.6%	56 of 145
How verbs show who acts and who receives the action	Grammar	59.5%	44 of 74
Types of sentences	Grammar	44.7%	38 of 85
Questions combining several grammar topics	Grammar	34.1%	75 of 220
Sound patterns and changes	Grammar	36.4%	28 of 77
Spelling	Grammar	57.1%	76 of 133
Punctuation	Grammar	49.6%	58 of 117
Problems in sentence wording	Grammar	64.6%	113 of 175

Confidence

Answers kept at 99% or more: 4, of which 0 wrong (share right: 51.0% to 100.0%, 95% interval); 2,194 skipped, the model less than 99% sure
Calibration gap: 21.0 points
AUROC, all questions: 0.73 0.71 to 0.75
AUROC, reading: 0.78 0.73 to 0.82
AUROC, grammar: 0.67 0.64 to 0.70
Brier score: 0.254
Accuracy of its surest answers: most confident fifth: 84.2% right; most confident half: 74.9% right; most confident four fifths: 64.6% right
Typical score of a right answer: 85.0%
Typical score of a wrong answer: 70.0%

Repeats

We did not repeat this setup.

Latency

Latency, median request: 1.07 s (this run)
Latency, 90th percentile: 1.22 s
Requests at once: 1
Throughput: 879.5 questions per minute of request time
Request time: 2.5 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: 253
Output tokens per question: 19