Qwen3.6-27B

by Qwen

Dense 27B multimodal model with flagship-level coding and vision

2.4Mdl/month

27Bparams

HuggingFace Use in Pipeline

Identifiers

Model ID

Qwen/Qwen3.6-27B

Feature URI

mixpeek://image_extractor@v1/qwen36_27b_v1

Overview

Qwen3.6-27B is Alibaba's dense 27-billion-parameter multimodal model that supports vision-language thinking and non-thinking modes in a single unified checkpoint. Despite being a dense model, it surpasses the previous 397B MoE flagship (Qwen3.5-397B-A17B) on every major coding benchmark and delivers strong vision understanding.

On Mixpeek, Qwen3.6-27B is the most powerful open-source captioning and visual reasoning model available, ideal for complex scene understanding, code extraction from screenshots, and detailed document analysis where accuracy matters more than throughput.

Architecture

64-layer dense language model using a hybrid layout of 16 repeats of (3x Gated DeltaNet + FFN, 1x Gated Attention + FFN) with hidden dim 5120 and FFN intermediate 17408. Supports 262K native context extensible to ~1M via YaRN. Trained with multi-token prediction.

Mixpeek SDK Integration

import { Mixpeek } from "mixpeek";

const mx = new Mixpeek({ apiKey: "API_KEY" });

await mx.collections.ingest({
  collection_id: "my-collection",
  source: { url: "https://example.com/video.mp4" },
  feature_extractors: [{
    name: "scene_description",
    version: "v1",
    params: {
      model_id: "Qwen/Qwen3.6-27B"
    }
  }]
});

Capabilities

Vision-language thinking and non-thinking modes in one checkpoint
262K native context window (extensible to ~1M tokens)
Flagship-level agentic coding (SWE-bench Verified: 77.2)
Strong visual understanding (MMMU: 82.9, VideoMME: 87.7)
Fits on a single consumer GPU with Q4_K_M quantization (16.8 GB)

Use Cases on Mixpeek

Complex visual scene analysis requiring deep reasoning across video content

Code extraction and understanding from screenshots and technical documentation

High-accuracy document analysis for legal, financial, and scientific content

Benchmarks

Dataset	Metric	Score	Source
MMMU	Accuracy	82.9%	Qwen3.6-27B blog post, April 2026
SWE-bench Verified	Resolve Rate	77.2%	Qwen3.6-27B blog post, April 2026
GPQA Diamond	Accuracy	87.8%	Qwen3.6-27B blog post, April 2026