Decide 9x faster than TypeSafe Jev 1.13, 33% cheaper

Ask a yes-or-no question of a message, a review or a comment: is it spam, is it negative. SIE, an open-source inference engine, serves DeBERTa v3 large zero-shot v2.0, a small open model that returns the probability of yes in one call: you state what yes means, with no training or fine-tuning.

Get started Run in your cloud

SIE DeBERTa v3 large zero-shot v2.0

Yes: spam

Text message SMS Spam Collection, CC BY 4.0

SIX chances to win CASH! From 100 to 20,000 pounds txt> CSH11 and send to 87575. Cost 150p/day, 6days, 16+ TsandCs apply Reply HL 4 info

What yes means

This text message is spam.

Probability of yes 0.99

TypeSafe Jev 1.13 keeps your app waiting 125 ms for each yes or no; GPT‑6 Luna, 884 ms

TypeSafe Jev 1.13

Over 100 ms

Asked

Is this comment on a music video spam? (with its rule for yes and no)

Wait for each decision 125 ms

100 ms to feel instant

GPT-6 Luna (short)

Over 100 ms

Asked

Is this comment spam?

Wait for each decision 884 ms

100 ms to feel instant

Nielsen Norman Group Nielsen: a response feels instant only within 0.1 second
Response Times: The 3 Important Limits

0.1 second is about the limit for having the user feel that the system is reacting instantaneously.
1 January 1993
arXiv:2608.20371 Study: Claude Haiku 4.5 took 981 ms and $0.25 per 1,000 calls to label one request
When Do LLMs Replace Fine-Tuned NLU? A Decision Framework for Intent Detection in Production Conversational Systems

Rodrigues and Vas

“the LLM’s median is 981 ms, with a p95 of 1,787 ms, above the 500 ms conversational bar.”
19 June 2026

9x faster than TypeSafe Jev 1.13 33% cheaper, at 95% of its accuracy

Run this eval

Decided correctly

1,560 comments ↑

TypeSafe Jev 1.13

TypeSafe Jev 1.13 95.47%

GPT-6 Luna (no reasoning)

GPT-6 Luna (no reasoning) 95.28%

Claude Haiku 5.5

Claude Haiku 5.5 93.10%

SIE DeBERTa v3 large zero-shot

SIE DeBERTa v3 large zero-shot 90.70%

TypeSafe Jev 1.13 (short)

TypeSafe Jev 1.13 (short) 88.06%

GPT-6 Luna (short)

GPT-6 Luna (short) 87.56%

Time per decision

median ↓

SIE DeBERTa v3 large zero-shot

13.8 ms SIE DeBERTa v3 large zero-shot

TypeSafe Jev 1.13

125 ms TypeSafe Jev 1.13

TypeSafe Jev 1.13 (short)

127 ms TypeSafe Jev 1.13 (short)

Claude Haiku 5.5

Claude Haiku 5.5 714 ms

GPT-6 Luna (short)

GPT-6 Luna (short) 884 ms

GPT-6 Luna (no reasoning)

GPT-6 Luna (no reasoning) 900 ms

Cost per 1,000 decisions

USD ↓

GPT-6 Luna (short)

$0.0074 GPT-6 Luna (short)

SIE DeBERTa v3 large zero-shot

$0.0084 SIE DeBERTa v3 large zero-shot

TypeSafe Jev 1.13 (short)

$0.0126 TypeSafe Jev 1.13 (short)

GPT-6 Luna (no reasoning)

$0.0140 GPT-6 Luna (no reasoning)

TypeSafe Jev 1.13

$0.0180 TypeSafe Jev 1.13

Claude Haiku 5.5

$0.0182 Claude Haiku 5.5

Tested: 1,560 YouTube comments, spam or not: the hardest of 3 public test sets
Scored: Right when it matches the dataset's label, at a cut-off set on 396 labelled comments
Prices: TypeSafe, OpenAI and Anthropic list prices; SIE at its Serverless price
Timing: SIE timed on a GPU server with the client alongside; providers over their public APIs

State what yes means, get its probability back

from sie_sdk import SIEClient
client = SIEClient(api_key="API keysk-sie-…")
result = client.extract(
    "modelMoritzLaurer/deberta-v3-large-zeroshot-v2.0",
    {"text": "textSIX chances to win CASH! From 100 to 20,000 pounds txt> CSH11 and send to 87575. Cost 150p/day, 6days, 16+ TsandCs apply Reply HL 4 info"},
    labels=[
        "labelThis text message is spam.",
    ],
    options={"hypothesis_template":"{}","multi_label":True},
)
print(result["classifications"])

import { SIEClient } from '@superlinked/sie-sdk';

const client = new SIEClient(undefined, { apiKey: 'sk-sie-…' });

const result = await client.extract(
  'MoritzLaurer/deberta-v3-large-zeroshot-v2.0',
  { text: "SIX chances to win CASH! From 100 to 20,000 pounds txt> CSH11 and send to 87575. Cost 150p/day, 6days, 16+ TsandCs apply Reply HL 4 info" },
  { labels: ["This text message is spam."], adapterOptions: {"hypothesis_template":"{}","multi_label":true} },
);
console.log(result.classifications);

curl https://api.superlinked.com/v1/extract/MoritzLaurer%2Fdeberta-v3-large-zeroshot-v2.0 \
  -H "Authorization: Bearer sk-sie-…" \
  -H "Content-Type: application/json" \
  -d "{\"items\":[{\"text\":\"SIX chances to win CASH"'!'" From 100 to 20,000 pounds txt> CSH11 and send to 87575. Cost 150p/day, 6days, 16+ TsandCs apply Reply HL 4 info\"}],\"params\":{\"labels\":[\"This text message is spam.\"],\"options\":{\"hypothesis_template\":\"{}\",\"multi_label\":true}}}"

Build the "Decide yes or no" capability into my app using the Superlinked Inference Engine (SIE).

Context
- SIE is an OpenAI-style inference API. Python SDK: `from sie_sdk import SIEClient`; TypeScript: `@superlinked/sie-sdk`.
- Base URL: https://api.superlinked.com (or my regional endpoint). Auth: Bearer key from env `SIE_API_KEY` (never hard-code it).
- Model: MoritzLaurer/deberta-v3-large-zeroshot-v2.0 (SIE primitive: /extract). Keep the model id configurable.

Task
- Input: a block of text plus a set of candidate labels.
- Behaviour: return the best-matching label(s) with native scores
- Call the selected SIE primitive once per request and map the response into your domain type.

Deliverables
- A typed client wrapper, an application-level function for this task, error handling for timeouts/empty input, and unit tests with a stubbed client.
- Wire it into my existing stack (ask me which framework if unclear) and add a short usage example.

Run

Output

classifications[0].label

This text message is spam.

classifications[0].score

0.9999

Make every yes-or-no call on Serverless or beside your app

Serverless API

We host the models; you call one API.

No idle GPUs, pay for what you use
Fits your stack: SDK, API, CLI, MCP
Self-host the same open-source engine
SOC 2 Type 2, US or EU data residency
Create an account to try Serverless

Get started

Self-hosted (Kubernetes)

Your data stays in your cloud, backed by our engineers.

Terraform modules for your cloud
Apache-2.0, same engine as Serverless
GPU workers scale to zero between jobs
Separate worker pools for each tenant

Agent prompt Deploy SIE to our AWS account with the superlinked/sie/aws Terraform module. Docs: superlinked.com/docs/deployment.

Deploy SIE to our Google Cloud project with the superlinked/sie/google Terraform module. Docs: superlinked.com/docs/deployment.

Deploy SIE to our Azure subscription with the superlinked/sie/azure Terraform module. Docs: superlinked.com/docs/deployment.

Deploy SIE to our Alibaba Cloud account with the superlinked/sie/alicloud Terraform module. Docs: superlinked.com/docs/deployment.

Talk to us Deploy guide

Run locally

The open-source server on your own machine.

NVIDIA GPU
One container, no cluster
Process your data fully offline
Same SDK and model IDs as Serverless

docker run --gpus all -p 8080:8080 ghcr.io/superlinked/sie-server:latest-cuda12-default

Quickstart

Switch to SIE in 5 minutes

from typesafe_sdk import Noul, TypeSafeClient

with TypeSafeClient(model="jev-latest") as client:
    response = client.system_one(
        state=message,
        questions={"spam": Noul(instructions=question)},
    )
p_yes = response.nouls["spam"].noul

from openai import OpenAI

client = OpenAI()
decision = client.decisions.create(
    model="gpt-6-luna",
    input=message,
    questions=[{"type": "predicate", "instructions": question}],
)
p_yes = decision.answers[0].probability

SIE

from sie_sdk import SIEClient

client = SIEClient(api_key="sk-sie-…")
result = client.extract(
    "MoritzLaurer/deberta-v3-large-zeroshot-v2.0",
    {"text": message},
    labels=[statement],
    options={"hypothesis_template":"{}","multi_label":True},
)
p_yes = result["classifications"][0]["score"]

from sie_sdk import SIEClient

client = SIEClient(api_key="sk-sie-…")
result = client.extract(
    "MoritzLaurer/deberta-v3-large-zeroshot-v2.0",
    {"text": message},
    labels=[statement],
    options={"hypothesis_template":"{}","multi_label":True},
)
p_yes = result["classifications"][0]["score"]