Decision model

StartLux-Decision-v1.1-9B-Instant

1 question per request

Accuracy, all 2,198 questions
51.2%49.1% to 53.3%, 95% interval; 1,126 right
Answers kept at 99% confidence
20 of 2,1980 wrong; share right 83.9% to 100.0%, 95% interval
Cost, $ per 1,000 questionsper 1M questions
No list priceBasis and price under Cost
Latency
0.154 smedian request, 1 question per request

Back to the leaderboard.

How we called it

Kind
Decision model
Access
StartLux-Decision server (startlux-decide on vLLM 0.31.0, the maker's release code), self-hosted, BF16
Paid as
Our own hardware
Model ID
startlux-models/StartLux-Decision-v1.1-9B-Instant
Parameters
9B dense, as the maker states it (counted from the released files: 9.65B).the whole checkpoint, its vision encoder included; source: counted from the checkpoint's safetensors, startlux-models/StartLux-Decision-v1.1-9B-Instant@18a17f7, huggingface.co/startlux-models/StartLux-Decision-v1.1-9B-Instant; read 10 November 2026
Hardware
1 × DGX Spark, GB10 (128 GB unified memory); BF16
Note
StartLux Labs' released StartLux-Decision-v1.1-9B-Instant (at 18a17f7), a fine-tune of Qwen3.5-9B that reads the prompt bidirectionally and also reads images. It ran on our hardware (one DGX Spark), one request at a time, through the server of its package (startlux-decide serve --launch-vllm, StartLux-Decision at a99977d) on vLLM 0.31.0, with vision on, as its 4B sibling. Its weights are licensed for non-commercial use only (CC BY-NC 4.0); we use them non-commercially, for evaluation. Its card lists English only.
Confidence from
Option score
Run
1290, 15 November 2026

Settings

Reasoning
off
Request format
1 question per request
Requests at once
1
Temperature
not recorded
Run record: harness seed, harness commit and prompt version
Harness seed
0
Harness commit
88e48d7a1055
Prompt version
254597b8be16

Results

QuestionsAccuracy95% intervalRight
All questions51.2%49.1% to 53.3%1,126 of 2,198
Reading78.8%75.5% to 81.8%517 of 656
Grammar39.5%37.1% to 42.0%609 of 1,542

Comparison group: αβγδζη. Compare with another setup.

Accuracy by topic
TopicPartAccuracyRight
Word and phrase meaningsReading67.9%89 of 131
Sentence meaningsReading86.4%133 of 154
How a writer explains and develops ideasReading71.4%50 of 70
Topic and main point of a paragraphReading92.9%91 of 98
How a paragraph fits togetherReading70.8%75 of 106
Details that support a paragraph’s main pointReading81.4%79 of 97
Word types, such as nouns and verbsGrammar47.9%67 of 140
Groups of words built around a nounGrammar32.3%21 of 65
Verbs and their formsGrammar31.9%29 of 91
Verb forms used like nouns or modifiersGrammar41.7%30 of 72
Word endings and what they doGrammar48.1%38 of 79
How words are builtGrammar31.9%22 of 69
The roles of words and phrases in a sentenceGrammar36.6%53 of 145
How verbs show who acts and who receives the actionGrammar52.7%39 of 74
Types of sentencesGrammar36.5%31 of 85
Questions combining several grammar topicsGrammar33.2%73 of 220
Sound patterns and changesGrammar23.4%18 of 77
SpellingGrammar33.1%44 of 133
PunctuationGrammar43.6%51 of 117
Problems in sentence wordingGrammar53.1%93 of 175

English

The Text tab's 547 kept reading questions translated into English, asked with this setup's Text tab settings. Every setup: English reading; against Turkish: English and Turkish.

Right answers
475 of 547, first English run (1293)
Accuracy
86.8%
Non-hallucination with “I don't know”
2.6%: 2 “I don't know” of 76 not right (run 1295)
Answered with “I don't know”
99.6%: 545 of 547
Error rate with “I don't know”
13.6%: 74 of 545 answers
Cost, $ per 1,000 questionsper 1M questions, first English run
No list price
Latency, median request, first English run
0.156 s; 1 request at once
English run
1293
Machine, English runs
DGX Spark
Date
15 November 2026
Request format
1 question per request
Reasoning
off
Prompt version
47f5776e71db
Translations, SHA-256
bf0d47cd385a93050c65906977e199ef7e45d5848dbbdef2bc8d124f4d5abf68

Confidence

Answers kept at 99% or more
20, of which 0 wrong (share right: 83.9% to 100.0%, 95% interval); 2,178 skipped, the model less than 99% sure
Calibration gap
7.6 points
AUROC, all questions
0.78 0.76 to 0.80
AUROC, reading
0.85 0.82 to 0.88
AUROC, grammar
0.69 0.66 to 0.72
Brier score
0.192
Accuracy of its surest answers
most confident fifth: 91.6% right; most confident half: 72.4% right; most confident four fifths: 57.7% right
Typical score of a right answer
75.6%
Typical score of a wrong answer
42.1%

Repeats

We did not repeat this setup.

Latency

Latency, median request
0.154 s (this run)
Latency, 90th percentile
0.169 s
Requests at once
1
Throughput
383.0 questions per minute of request time
Request time
5.7 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

Cost per 1,000 questionsper 1M questions
No list price
Basis
No dollar estimate; see hardware and time
Input tokens per question
363
Output tokens per question
0