Alibaba

Qwen3.8 27B

About

Qwen3.8 27B is a 27B open-weight model from Alibaba. Reasoning VLM (image + video + text). It accepts text, images, and video. Context window is 164K. Released August 13, 2026.

Parameters
27B
Published parameter count.
Context window
164K
Tokens of context on a request.
License
Apache-2.0
License on the weights.

Pricing

Token rates are US dollars per 1M tokens. Image rates are per 1K images. Audio rates are per 1M audio seconds.

Rate	Price
Input / 1M	$0.35
Cached input / 1M	$0.05
Output / 1M	$2.55

Compare

Qwen3.8 27B is one of the catalog's chat models. The table is each model's published price, size, and context.

Model	What it does	Price	Parameters	Context
Qwen3.8 27B	Reasoning VLM (image + video + text)	$0.35 / 1M input	27B	164K
Qwen3.5 0.8B	Multimodal chat model with up to 64 images or one video per request.	$0.08 / 1M input	0.8B	256K
Gemini 3.5 Flash Lite	Fastest and cheapest Gemini tier. Multimodal chat; emits no reasoning tokens.	$0.30 / 1M input	—	1M
Gemini 3.7 Flash	Newest Gemini Flash tier with extended reasoning. Multimodal chat + native video.	$0.75 / 1M input	—	1M

Benchmarks

Published scores for Qwen3.8 27B, from the Qwen3.8 model card.

Evals

Reading	Example
Qualitative	Clear structure, grounded in the input
Quantitative	OmniDocBench 1.5: 91.1.
Cost and performance	Lower listed rate, mid-pack latency

Performance

Latency is the end-to-end round trip. Throughput is completion tokens divided by that time. It is a request rate, not decode speed.

Latency
2,203 ms
Throughput
384 tok/s
Requests
32,190

Supported inputs

textYes
Text-only prompts are accepted.
image_urlYes
Up to 64 per request, read together as one prompt. See Image Inputs.
video_urlYes
One clip. Tune sampling with video_fps, video_max_frames, and video_resolution at the top level. See Video Inputs.
document_urlNo
A document_url part is a 400 capability error. Use an OCR model such as paddleocr/pp-ocrv6 for PDFs.

Default and only method: free-form chat (no method or method_params). response_format={"type":"json_schema"} is supported.
Context window 256K tokens. On the catalog it reports provider: "vlm-run".
Weights: Inferact/Qwen3.8-27B-NVFP4, based on upstream Qwen/Qwen3.8-27B. Alias: Qwen/Qwen3.8-27B.

Methods

Methods this model serves. Payload shapes are in the docs.

Method	Returns
chat (default)	The model's reply, verbatim

A chat VLM is pass-through. Text mode returns the reply with no wrapper, even with several images: the model reads them together and produces one reply. JSON mode returns the model's own JSON, with no data envelope. See Methods & Response Format.

Estimate cost

Estimate. An hour of video is 15 frames a minute at 256 tokens a frame, plus the audio in that request, and 500 output tokens a minute.

Hours of video

$197.46

Quick start

Call this model on the OpenAI-compatible gateway. The model id is already filled in.

from openai import OpenAI

client = OpenAI(
    base_url="https://gateway.vlm.run/v1/openai",
    api_key="<VLMRUN_API_KEY>",
)

response = client.chat.completions.create(
    model="qwen/qwen3.8-27b",
    messages=[
        {
            "role": "user",
            "content": [
                {"type": "text", "text": "What is happening in this image?"},
                {
                    "type": "image_url",
                    "image_url": {
                        "url": "https://storage.googleapis.com/vlm-data-public-prod/hub/examples/image.caption/car.jpg"
                    },
                },
            ],
        }
    ],
)

print(response.choices[0].message.content)

Response

A silver sedan is parked on a sunlit street beside a row of trees.

For video, pass a video_url part and tune sampling with video_fps / video_max_frames. See Video Inputs.