Chat model

DeepSeek-V4.1-Flash, first release

reasoning off · 16 questions per request

65.9% of the 2,198 questions right (63.9% to 67.9%, 95% interval). Back to the leaderboard.

How we called it

Kind: Chat model
Access: SGLang, self-hosted, published mixed precision; first-release weights (revision fb2764a), not the 1017 update
Paid as: Our own hardware
Model ID: deepseek-v4.1-flash
Parameters: 552B (8 to 16B active per token)
Hardware: 4 × RTX PRO 6000 Blackwell (96 GB); FP4 experts, FP8 and BF16 layers
Confidence from: Answer letter probability
Run: 120, 5 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	65.9%	63.9% to 67.9%	1,449 of 2,198
Reading	87.8%	85.1% to 90.1%	576 of 656
Grammar	56.6%	54.1% to 59.1%	873 of 1,542

Comparison group: pqrs. Compare with another setup.

Accuracy by topic

Topic	Part	Accuracy	Right
Word and phrase meanings	Reading	84.7%	111 of 131
Sentence meanings	Reading	90.9%	140 of 154
How a writer explains and develops ideas	Reading	80.0%	56 of 70
Topic and main point of a paragraph	Reading	95.9%	94 of 98
How a paragraph fits together	Reading	80.2%	85 of 106
Details that support a paragraph’s main point	Reading	92.8%	90 of 97
Word types, such as nouns and verbs	Grammar	60.0%	84 of 140
Groups of words built around a noun	Grammar	52.3%	34 of 65
Verbs and their forms	Grammar	65.9%	60 of 91
Verb forms used like nouns or modifiers	Grammar	63.9%	46 of 72
Word endings and what they do	Grammar	58.2%	46 of 79
How words are built	Grammar	40.6%	28 of 69
The roles of words and phrases in a sentence	Grammar	47.6%	69 of 145
How verbs show who acts and who receives the action	Grammar	71.6%	53 of 74
Types of sentences	Grammar	54.1%	46 of 85
Questions combining several grammar topics	Grammar	37.3%	82 of 220
Sound patterns and changes	Grammar	40.3%	31 of 77
Spelling	Grammar	76.7%	102 of 133
Punctuation	Grammar	55.6%	65 of 117
Problems in sentence wording	Grammar	72.6%	127 of 175

Confidence

Answers kept at 99% or more: 576, of which 5 wrong (share right: 98.0% to 99.6%, 95% interval); 1,622 skipped, the model less than 99% sure
Calibration gap: 13.5 points
AUROC, all questions: 0.82 0.81 to 0.84
AUROC, reading: 0.91 0.88 to 0.93
AUROC, grammar: 0.76 0.73 to 0.78
Brier score: 0.187
Accuracy of its surest answers: most confident fifth: 99.5% right; most confident half: 88.6% right; most confident four fifths: 74.2% right
Typical score of a right answer: 97.1%
Typical score of a wrong answer: 62.0%

Repeats

Run 5 times on a sample of 200 questions.

Answers changed in 5 runs: 12 of 200
Same wrong answer in all 5: 63 of 200
Wrong in at least one run: 75 of 200
Accuracy in each run: 64.5%, 64.0%, 64.5%, 64.0%, 64.5%

Latency

Latency, median request: 4.36 s (this run)
Latency, 90th percentile: 7.86 s
Requests at once: 1
Throughput: 177.5 questions per minute of request time
Request time: 12.4 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: 303
Output tokens per question: 6