Answer image questions 91% cheaper than GPT-⁠6.1 Sol

SIE is the open-source inference engine serving Qwen3.8 27B. The model answers questions about your photos, or says when the answer is not visible.

Get started Run in your cloud

SIE Qwen3.8 27B

9 of 10 people agree

Question from a blind userWhat kind of coffee is this?

AnswerFolgers

VizWiz · CC BY 4.0

GPT-⁠6.1 Sol and Claude Sonnet 5.5 answer what the photo doesn’t show

The blue lid of a yogurt tub on a floor next to a pair of jeans; no date is printed on the visible part — VizWiz · CC BY 4.0

Question from a blind userWhat is the expire date on this yogurt?

10 of 10 people said the photo doesn’t show it

GPT-6.1 Sol (medium reasoning)

Gave a date that isn’t there

17/03

GPT-6 Luna (high reasoning)

Gave a date that isn’t there

12/14

A completely blurred close-up of a bottle label with a few dark stripes of unreadable text and an orange edge — VizWiz · CC BY 4.0

Question from a blind userWhat is in this bottle?

9 of 10 people said the photo doesn’t show it

Claude Sonnet 5.5 (adaptive thinking)

Guessed what’s inside

Prescription medication (pills).

GPT-6.1 Sol (medium reasoning)

Guessed what’s inside

Orange juice

arXiv:2602.13469 CHI 2026 A GPT-4o vision app gave blind users false information in 1 of 5 follow-up answers
How Multimodal Large Language Models Support Access to Visual Information: A Diary Study With Blind and Low Vision People

Gonzalez Penuela, Jung, Lin, Hu and Azenkot

“the AI did not abstain when the text was unreadable but still attempted to provide confident-sounding answers.”
13 February 2026
arXiv:2604.14799 ServiceNow Research Stronger abstention prompts still leave GPT-5 and Claude Sonnet 4.5 guessing
Knowing When Not to Answer: Evaluating Abstention in Multimodal Reasoning Systems

Madhusudhan, Yadav and Lacoste
16 April 2026

41% fewer missed abstentions than GPT-⁠6.1 Sol, 91% cheaper

Run this eval

Required abstentions missed

299 unanswerable photos ↓

SIE Qwen3.8 27B

SIE Qwen3.8 27B 47.5%

GPT-6 Luna (medium reasoning)

GPT-6 Luna (medium reasoning) 70.9%

Claude Haiku 5.5

Claude Haiku 5.5 79.3%

GPT-6.1 Sol

GPT-6.1 Sol 80.6%

Claude Sonnet 5.5

Claude Sonnet 5.5 84.6%

Cost per 1,000 questions

USD ↓

GPT-6 Luna (medium reasoning)

$0.289 GPT-6 Luna (medium reasoning)

Claude Haiku 5.5

$0.307 Claude Haiku 5.5

SIE Qwen3.8 27B

$0.425 SIE Qwen3.8 27B

GPT-6.1 Sol

GPT-6.1 Sol $4.75

Claude Sonnet 5.5

Claude Sonnet 5.5 $5.19

Evaluation

Tested: 600 VizWiz photo questions; 299 crowd-labelled unanswerable
Scored: Exact “unanswerable” reply; longer abstentions and failures count as misses
Prices: SIE token-mix estimate; rival list-price usage, 11 October 2026

Send an image and a question, get the answer

from pathlib import Path
from sie_sdk import SIEClient
client = SIEClient(api_key="API keysk-sie-…")
image = {"data": Path(imageVizWiz_test_00000984.jpgbrowse).read_bytes(), "format": "jpeg"}
result = client.generate(
    "modelQwen/Qwen3.8-27B-FP8",
    "queryWhat kind of coffee is this?
Answer the question using a single word or phrase. Give your best answer from what is visible, even if the photo is blurry or shows only part of it.
Reply 'Unanswerable' only if the answer cannot be seen in the photo at all.",
    max_new_tokens=128,
    images=[image],
    temperature=0,
    presence_penalty=0,
)
print(result["text"])

import { readFile } from 'node:fs/promises';
import { SIEClient } from '@superlinked/sie-sdk';

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

const image = await readFile("VizWiz_test_00000984.jpg");
const result = await client.generate(
  'Qwen/Qwen3.8-27B-FP8',
  "What kind of coffee is this?\nAnswer the question using a single word or phrase. Give your best answer from what is visible, even if the photo is blurry or shows only part of it.\nReply 'Unanswerable' only if the answer cannot be seen in the photo at all.",
  { maxNewTokens: 128, images: [image], temperature: 0, presencePenalty: 0 },
);
console.log(result.text);

image_bytes=$(base64 < 'VizWiz_test_00000984.jpg' | tr -d '\n')
curl https://api.superlinked.com/v1/generate/Qwen__Qwen3.8-27B-FP8 \
  -H "Authorization: Bearer sk-sie-…" \
  -H "Content-Type: application/json" \
  -d "{\"prompt\":\"What kind of coffee is this?\\nAnswer the question using a single word or phrase. Give your best answer from what is visible, even if the photo is blurry or shows only part of it.\\nReply 'Unanswerable' only if the answer cannot be seen in the photo at all.\",\"max_new_tokens\":128,\"images\":[{\"data\":\"$image_bytes\",\"format\":\"jpeg\"}],\"temperature\":0,\"presence_penalty\":0}"

Build the "Image questions" 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: Qwen/Qwen3.8-27B-FP8 (SIE primitive: /generate). Keep the model id configurable.

Task
- Input: an uploaded image plus a question or caption instruction.
- Behaviour: return a detailed caption or answer the question about the image
- Send one POST https://api.superlinked.com/v1/generate/Qwen__Qwen3.8-27B-FP8 request (SDK: generate) with the question as prompt, the image in images, max_new_tokens 128, temperature 0 and presence_penalty 0, so answers are greedy and repeatable. Read the answer from text.

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

A hand holding a red Folgers Gourmet Supreme coffee can — VizWiz_test_00000984.jpg

text Folgers

Answer image questions with SIE

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
Zero lock-in, self-host the same stack
SOC 2 Type 2, US or EU data residency
$50 of credit to start, no card required

Get started

Self-hosted (Kubernetes)

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

Terraform to your cloud in minutes
Apache-2.0, same engine as Serverless
GPU workers scale to zero between jobs
Per-tenant pools, no noisy neighbors

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 H100 80GB
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-sglang

Quickstart

Switch to SIE in 5 minutes

from openai import OpenAI

client = OpenAI(api_key="sk-…")
reply = client.responses.create(
    model="gpt-6-luna",
    input=[{"role": "user", "content": [
        {"type": "input_image", "image_url": image_data_url},
        {"type": "input_text", "text": prompt},
    ]}],
)
answer = reply.output_text

import anthropic

client = anthropic.Anthropic(api_key="sk-ant-…")
reply = client.messages.create(
    model="claude-haiku-5-5",
    max_tokens=1024,
    messages=[{"role": "user", "content": [
        {"type": "image", "source": {"type": "base64", "media_type": "image/jpeg", "data": image_b64}},
        {"type": "text", "text": prompt},
    ]}],
)
answer = next(block.text for block in reply.content if block.type == "text")

SIE

from sie_sdk import SIEClient

client = SIEClient(api_key="sk-sie-…")
reply = client.generate("Qwen/Qwen3.8-27B-FP8", prompt, images=[image])
answer = reply["text"]

from sie_sdk import SIEClient

client = SIEClient(api_key="sk-sie-…")
reply = client.generate("Qwen/Qwen3.8-27B-FP8", prompt, images=[image])
answer = reply["text"]