Qwen3.8-27B is Alibaba Cloud's latest open-weight, dense vision-language model. It accepts text, image, and video input and is built for coding, professional workflows, research, multimodal interaction, and long-running agent tasks, with reasoning effort that can be tuned per request.
| Specification | Details |
|---|---|
| Developer | Alibaba Cloud (Qwen) |
| Model ID | alibaba/qwen3.8-27b |
| Release Date | August 14, 2026 |
| Architecture | Dense, open-weight vision-language model |
| Context Window | 262,144 tokens (~256K) |
| Max Output Tokens | 131,072 |
| Modalities (input) | Text · Image · Video |
| Modalities (output) | Text |
| Capabilities | Vision, tunable reasoning, streaming, structured output |
Text, images, and video are all first-class inputs, so the model can be used for document QA, visual analysis, and video-aware tasks without bolting on a separate vision pipeline.
Reasoning effort can be adjusted low, medium, or high depending on the task — fast and cheap for simple queries, deeper and slower when a problem needs more careful multi-step thinking.
Responses can be returned as strict structured JSON or streamed token-by-token, making the model workable both for user-facing chat interfaces and for backend pipelines that need machine-readable output.
Native text, image, and video input means it can be used directly on screenshots, scanned documents, or video frames without a separate vision model in front of it.
Tunable reasoning effort lets you dial up depth for harder coding or research tasks, and dial it back down for quick, low-latency queries.
Built with agentic workflows in mind, making it a reasonable fit for multi-step tasks that call tools and reason over multimodal context along the way.
Measured on the AI/ML API gateway across live traffic. Figures are medians unless stated. Last updated 2026-10-12.
Qwen3.8 27B is documented as suited to coding and software engineering tasks, including repository-level code generation, refactoring, bug fixing, and multi-file edits.
It is also documented as suited to long-horizon agentic work, such as autonomous planning, multi-step task execution, and tool orchestration across long sessions.
The model is documented for multimodal vision-language use, including image and video understanding, OCR, visual question answering, and chart, diagram, and document understanding.
It is also documented for complex problem solving and professional work, including multi-step reasoning, mathematical problem solving, research, and decision support.
import requests r = requests.post( "https://api.aimlapi.com/v1/chat/completions", headers={"Authorization": "Bearer " + AIMLAPI_KEY}, json={ "model": "alibaba/qwen3.8-27b", "messages": [ { "role": "user", "content": "Hello!" } ] }, ) print(r.json())
const r = await fetch("https://api.aimlapi.com/v1/chat/completions", { method: "POST", headers: { Authorization: `Bearer ${process.env.AIMLAPI_KEY}`, "Content-Type": "application/json", }, body: JSON.stringify({ "model": "alibaba/qwen3.8-27b", "messages": [ { "role": "user", "content": "Hello!" } ] }), }); console.log(await r.json());
curl -X POST https://api.aimlapi.com/v1/chat/completions \ -H "Authorization: Bearer $AIMLAPI_KEY" \ -H "Content-Type: application/json" \ -d '{"model":"alibaba/qwen3.8-27b","messages":[{"role":"user","content":"Hello!"}]}'
OpenAI-compatible — swap the base URL and it works with your existing SDK.
| Type | Price |
|---|---|
| Input | |
| Output | |
| Benchmark | Score | What it measures | Source | Retrieved |
|---|---|---|---|---|
| Intelligence | 33.9 | Composite score across standardised reasoning, knowledge and problem-solving evaluations, measured independently by Artificial Analysis | Source | September 12, 2026 |
| Coding | 68.1 | Composite score across standardised coding evaluations, measured independently by Artificial Analysis | Source | September 12, 2026 |
| Terminal-Bench | 73.0 | Autonomous shell/terminal task completion | Source | September 22, 2026 |
| LiveCodeBench | 90.3 | Contamination-free competitive programming problems | Source | September 22, 2026 |
| OSWorld | 84.3 | Computer-use across real desktop applications | Source | September 22, 2026 |
| GPQA Diamond | 89.2 | Google-proof graduate science questions (hardest subset) | Source | September 22, 2026 |
| Humanity's Last Exam | 30.8 | Expert-level questions across many domains | Source | September 22, 2026 |
| Model | Input | Output | Context | Best for |
|---|---|---|---|---|
Qwen 3.8 27B This page | Multimodal + agents | |||
| Reasoning + agents | ||||
| Balanced coding + agents | ||||
| Long-context, multimodal & agentic workflows | ||||
| Reasoning + agents |
Qwen 3.8 27B has a 262,144 tokens context window and can return up to 131,072 tokens.
Qwen 3.8 27B takes image, text, video as input and returns text.
Use alibaba/qwen3.8-27b as the model id. Requests go to https://api.aimlapi.com/v1/chat/completions.
Qwen 3.8 27B became available on August 14, 2026.
Qwen 3.8 27B is priced at input $0.61893 / 1M tokens, output $4.40128 / 1M tokens.
Yes, Qwen 3.8 27B can stream responses as they are generated.
Yes, Qwen 3.8 27B accepts image input alongside text.
Qwen 3.8 27B was built by Alibaba Cloud.
It is an open-weight dense vision-language chat model from Alibaba Cloud, designed for coding, research, multimodal interaction and long-running agent tasks.
Yes, it accepts image and video inputs in addition to text, and produces text output.
Yes, it includes a reasoning capability that can be tuned per request.
Yes, it supports structured output generation.
