Skip to main content

Supersonic LabsDecision modelOpen source

Julia 1

PopularityTop group about equally popular, 12 models

41out of 100new calculation

based on two data groups, mostly search, range 26 to 57

Summary

Julia 1 is Supersonic Labs’ first checkpoint in the Julia family, with 144.3 million parameters. The multilingual model processes a state, a question, and possible answers, then returns Choice, Noul, or Score. It fits classification and routing while keeping the readout structured instead of producing a free-form explanation.

It is based on jhu-clsp/mmBERT-small. Local PyTorch or ONNX execution supports up to 8,192 runtime tokens under an Apache-2.0 license.

GraduallyBench Decision

Julia 1 decides 17 of 60 test cases correctly. The median of the 33 tested decision models is 60.0 %.

Accuracy

28.3 %

17 of 60 test cases

Robustness

43.3 %

13 of 30 reworded variants

Unnecessary reviews

13.5 %

7 of 52 cases without a required review

Response time

42 ms

median, 95 % under 60 ms, local, CPU

Accuracy compared

Share of correct decisions in percent, higher is better

20 of 33 models

As of 11/10/2026. Source: GraduallyBench Decision

Accuracy by task familyThe tick marks the median of all 33 tested models.
Triage25.0 %3 / 12, median 58.3 %
Routing50.0 %6 / 12, median 83.3 %
Constraints16.7 %2 / 12, median 33.3 %
Tables25.0 %3 / 12, median 50.0 %
Evidence25.0 %3 / 12, median 75.0 %

Julia 1

Benchmark run
Show test details

Decisions

Local · German, English

Execution Not documented. Precision not documented.

130 valid outputs from 130 attempted and 130 planned test slots.

Measured duration p50 / p95 41.5 / 60 ms (130 cases). Local CPU. Quantile calculation not documented.

Automatic decision accuracy
28.33 %(17 / 60 planned cases)German, English
Automatic decision accuracy, Triage
25 %(3 / 12 planned cases)German, English
Automatic decision accuracy, Routing
50 %(6 / 12 planned cases)German, English
Automatic decision accuracy, Selection with constraints
16.67 %(2 / 12 planned cases)German, English
Automatic decision accuracy, Numbers and tables
25 %(3 / 12 planned cases)German, English
Automatic decision accuracy, Sufficient evidence
25 %(3 / 12 planned cases)German, English
Correct robustness variants
43.33 %(13 / 30 planned cases)German, English
Correct development cases
40 %(16 / 40 planned cases)German, English
Unnecessary reviews
13.46 %(7 / 52 assessable cases without a required review)German, English
Missed required reviews
75 %(6 / 8 assessable cases requiring review)German, English

Response times are comparable only within the same environment and form no shared speed ranking. Where used, the test references were not independently reviewed by humans. Test coverage and methods differ by model. Each metric stands on its own; there is no combined overall score across tests. Data compiled on 11/10/2026. How we test · Download all Decision and Audio results as JSON

All benchmark results

Every published result with its source, test setup and comparison models.

Show all 112 published benchmark results

Page 1 of 10

Measurements without matching peer values

These measurements have no matching peer values under the same test conditions. Their original values and sources remain available here.

Typed decisions: accuracy (Show measurement, test conditions, and source)
Source value
73.15
Metric
accuracy
Unit
%
Category
Typed decisions
Direction
Higher is better
Sample
2000
Source type
vendor-reported
Evaluator
SupersonicLabs Julia-1 model card
Status
active
Retrieved at
2026-10-04
Methodology
Vendor-reported result. Evaluated at=2026-09-24; Hardware=H200; Precision=BF16; Strict encoding=yes; Maximum token length=1024; Correct answers=1463; Sample size=2000
Source SupersonicLabs Julia-1 model card
AG News · 4 labels: accuracy (Show measurement, test conditions, and source)
Source value
94
Metric
accuracy
Unit
%
Category
AG News · 4 labels
Direction
Higher is better
Sample
100
Source type
vendor-reported
Evaluator
SupersonicLabs Julia-1 model card
Status
active
Retrieved at
2026-10-04
Methodology
Vendor-reported result. Evaluated at=2026-09-24; Hardware=H200; Precision=BF16; Strict encoding=yes; Maximum token length=1024; Correct answers=94; Sample size=100
Source SupersonicLabs Julia-1 model card
DAIR Emotion · 6 labels: accuracy (Show measurement, test conditions, and source)
Source value
86
Metric
accuracy
Unit
%
Category
DAIR Emotion · 6 labels
Direction
Higher is better
Sample
100
Source type
vendor-reported
Evaluator
SupersonicLabs Julia-1 model card
Status
active
Retrieved at
2026-10-04
Methodology
Vendor-reported result. Evaluated at=2026-09-24; Hardware=H200; Precision=BF16; Strict encoding=yes; Maximum token length=1024; Correct answers=86; Sample size=100
Source SupersonicLabs Julia-1 model card
Banking77 pilot · 72 labels: accuracy (Show measurement, test conditions, and source)
Source value
64
Metric
accuracy
Unit
%
Category
Banking77 pilot · 72 labels
Direction
Higher is better
Sample
100
Source type
vendor-reported
Evaluator
SupersonicLabs Julia-1 model card
Status
active
Retrieved at
2026-10-04
Methodology
Vendor-reported result. Evaluated at=2026-09-24; Hardware=H200; Precision=BF16; Strict encoding=yes; Maximum token length=1024; Correct answers=64; Sample size=100; Pilot=yes; Labels=72; Shortlist=ranking/top-16
Source SupersonicLabs Julia-1 model card
MASSIVE scenario classification: All 52 locales: scenario accuracy (Show measurement, test conditions, and source)
Source value
71.5
Metric
scenario accuracy
Unit
%
Category
MASSIVE scenario classification: All 52 locales
Direction
Higher is better
Sample
154648
Source type
vendor-reported
Evaluator
SupersonicLabs Julia-1 model card
Status
active
Retrieved at
2026-10-04
Methodology
Vendor-reported result. Labels=18; locales=52; Correct answers=110573; Sample size=154648; Not intent or slot filling=yes
Source SupersonicLabs Julia-1 model card
MASSIVE scenario classification: Portuguese · pt-PT: scenario accuracy (Show measurement, test conditions, and source)
Source value
86.25
Metric
scenario accuracy
Unit
%
Category
MASSIVE scenario classification: Portuguese · pt-PT
Direction
Higher is better
Sample
2974
Source type
vendor-reported
Evaluator
SupersonicLabs Julia-1 model card
Status
active
Retrieved at
2026-10-04
Methodology
Vendor-reported result. Labels=18; locales=1; Correct answers=2565; Sample size=2974; Not intent or slot filling=yes
Source SupersonicLabs Julia-1 model card
MASSIVE scenario classification: English · en-US: scenario accuracy (Show measurement, test conditions, and source)
Source value
86.75
Metric
scenario accuracy
Unit
%
Category
MASSIVE scenario classification: English · en-US
Direction
Higher is better
Sample
2974
Source type
vendor-reported
Evaluator
SupersonicLabs Julia-1 model card
Status
active
Retrieved at
2026-10-04
Methodology
Vendor-reported result. Labels=18; locales=1; Correct answers=2580; Sample size=2974; Not intent or slot filling=yes
Source SupersonicLabs Julia-1 model card
typed-decisions: Choice: accuracy (Show measurement, test conditions, and source)
Source value
71.33
Metric
accuracy
Unit
%
Category
typed-decisions: Choice
Direction
Higher is better
Sample
600
Source type
vendor-reported
Evaluator
SupersonicLabs Julia-1 model card
Status
active
Retrieved at
2026-10-04
Methodology
Vendor-reported result. Evaluated at=2026-09-24; Hardware=H200; Precision=BF16; Strict encoding=yes; Maximum token length=1024; Correct answers=428; Sample size=600
Source SupersonicLabs Julia-1 model card
typed-decisions: Noul: accuracy (Show measurement, test conditions, and source)
Source value
80.67
Metric
accuracy
Unit
%
Category
typed-decisions: Noul
Direction
Higher is better
Sample
600
Source type
vendor-reported
Evaluator
SupersonicLabs Julia-1 model card
Status
active
Retrieved at
2026-10-04
Methodology
Vendor-reported result. Evaluated at=2026-09-24; Hardware=H200; Precision=BF16; Strict encoding=yes; Maximum token length=1024; Correct answers=484; Sample size=600
Source SupersonicLabs Julia-1 model card
typed-decisions: Score: accuracy (Show measurement, test conditions, and source)
Source value
68.88
Metric
accuracy
Unit
%
Category
typed-decisions: Score
Direction
Higher is better
Sample
800
Source type
vendor-reported
Evaluator
SupersonicLabs Julia-1 model card
Status
active
Retrieved at
2026-10-04
Methodology
Vendor-reported result. Evaluated at=2026-09-24; Hardware=H200; Precision=BF16; Strict encoding=yes; Maximum token length=1024; Correct answers=551; Sample size=800
Source SupersonicLabs Julia-1 model card
DecideBench overall accuracy (Show measurement, test conditions, and source)
Source value
34.8
Metric
accuracy
Unit
%
Category
decision
Direction
Higher is better
Sample
400
Source type
independent-evaluator
Evaluator
DecideBench
Status
active
Retrieved at
2026-10-18
Methodology
Independent evaluation. DecideBench results read from the repository main README (not pinned to a commit); 400 multiple-choice decisions in 200 contrastive pairs, measured 2026-09-28 through 2026-10-09. Exact per-model checkpoint revision and request sampling settings are not reported. Exact row is Julia-1 144M (self-hosted) on an NVIDIA L4 at four requests in flight; the README explicitly says Julia-1 ran zero-shot. Exact checkpoint revision and inference settings are not further stated.
Source DecideBench
DecideBench pair accuracy (Show measurement, test conditions, and source)
Source value
8
Metric
pair accuracy
Unit
%
Category
decision
Direction
Higher is better
Sample
200
Source type
independent-evaluator
Evaluator
DecideBench
Status
active
Retrieved at
2026-10-18
Methodology
Independent evaluation. DecideBench results read from the repository main README (not pinned to a commit); 400 multiple-choice decisions in 200 contrastive pairs, measured 2026-09-28 through 2026-10-09. Exact per-model checkpoint revision and request sampling settings are not reported. Exact row is Julia-1 144M (self-hosted) on an NVIDIA L4 at four requests in flight; the README explicitly says Julia-1 ran zero-shot. Exact checkpoint revision and inference settings are not further stated.
Source DecideBench

Page 1 of 10

Popularity compared

How often the model is searched for, used and mentioned, compared with every model of its category.

Popularity index

0 to 100, higher is more popular

20 of 42 models

As of November 2026 · Source: gradually.ai popularity index

Over time

Weekly values from October 18, 2026 to November 20, 2026

Julia 1Jev 43Decider 2B 41Kev 0.8B 40Kev 9B 40

Method changes in the timeline, a jump at these points is no market event:

  1. 1November 1, 2026: Value only with search or usage data, 30-day attention
  2. 2November 1, 2026: Method version 4 (common scale, shrinkage)
  3. 3November 9, 2026: New topics measurement V20 (daily quorum, short forms)

What is gaining popularity right now, across models, vendors, harnesses and AI subscriptions, is collected in All AI trends.

All specifications

Architecture

Model class
Multilingual decision model for classification and routing, not a chat model
Source: Model class
Checkpoint
SupersonicLabs/Julia-1
Source: Checkpoint
Base model
jhu-clsp/mmBERT-small
Source: Base model

Input and output

Input
State, question, and possible answers as text
Source: Input
Context window
Up to 8,192 tokens at runtime
Source: Context window

Watermarking and provenance

AI watermarking refers to invisible signals embedded in the output. Visible marks and provenance metadata are listed separately. Unknown means there is insufficient evidence for that particular field.

AI watermarking
Unknown
no source
Watermarking technique
Not known
no source

Licence and access

Access
Local PyTorch or ONNX execution
Source: Access
License
Apache-2.0
Source: License
Shutdown
None announced
no source

Further details

Output schema
Choice, Noul, and Score over the options supplied by the caller
Source: Output schema

Sources and data date

Show all 13 sources

Benchmark sources at a glance

Which independent evaluators tested this model. This is coverage, not a quality ranking.

No active result from the reviewed external evaluators can currently be attributed to this model. Other source evidence remains available below.

Without vendor claims and historical or disputed results; those stay labelled in the single values.

TypeEvidence and data date
Research statusResearch date October 4, 2026. 7 source URLs checked. This documents the inspected sources, not an exhaustive inventory of every publication.
Unresolved

evaluation-configurations

Note

The H200 BF16 evaluation and the CPU FP32 reproduction are separate runs and are not interchangeable.

Not published

Immutable Decision Index served checkpoint revision

Note

The current public index gives a model URL and, where available, a served_checkpoint label, but no immutable commit/digest for the exact evaluated weights.

Additional sourceJulia 1 model card (retrieved October 3, 2026; October 4, 2026) · Julia 1 model card · Editorial description reviewed October 3, 2026
Additional sourceDecideBench (retrieved October 18, 2026)
Additional sourceSupersonicLabs Julia-1 metrics file (retrieved October 22, 2026)
Additional sourceSysOne-bench public results.json (retrieved October 23, 2026)
Additional sourceJev Decision Index public index.json (release-v4) (retrieved October 28, 2026)
Additional sourceModel metadata (retrieved October 18, 2026)
Additional sourceModel metadata (retrieved October 22, 2026)
Additional sourceOpenRouter: Popularity index (derived), CC BY 4.0
Additional sourceVercel AI Gateway: Popularity index (derived), CC BY 4.0
Additional sourceOllama: cumulative downloads for the popularity index
Additional sourceReplicate: cumulative runs for the popularity index
Additional sourceGoogle Trending Now: limited feed selection for Germany and the US
Additional sourceCreative Commons Attribution 4.0