Qwen3.8 Flash Next

Tool callingStructured outputReasoningImage inputVideo input

Open-weight experimental preview of the Qwen4 architecture: hybrid-attention MoE (125B total, 6B active) with vision encoder for coding, agent tasks, and image and video understanding

Context: 262K
Released: 1mo ago
Parameters: —

Sources:models.dev

Benchmarks

DeepSWE is the primary coding benchmark where an exact model evaluation exists. Other sourced evaluations appear below; missing scores are not zero.

Not evaluated in DeepSWE 1.1 as of 2026-09-22. Results are only shown for exact model versions in the official dataset. Official leaderboard.

No benchmark results have been matched to this model yet.

Specifications

Developer: Alibaba
Model family: Qwen
Type: Language model
Predecessor: —
Successor: —
Released: August 27, 2026
Model ID: alibaba/qwen3.8-flash-next
Openness: Open weights
License: qwen-community-1.0
Parameters: —
Input modalities: text, image, video
Output modalities: text
Modality: text+image+video->text
Max output tokens: 131K
Hugging Face: —
Source freshness: Metadata synced Oct 8, 2026, 10:17 PM UTC

Additional source evidence

Source model: alibaba/qwen3.8-flash-next
Source family: qwen
Last updated: 2026-08-27
Model context: 262K
Model output limit: 131K
Open weights: Yes
File attachments: Yes
Reasoning: Yes
Tool calling: Yes
Structured output: Yes
Temperature control: Yes

Downloadable weights

Hugging Face

Published benchmarks

Benchmark	Score	Evaluation details	Evidence
DeepSWE	58.7 · score	mini-swe-agent · 1.1	Source
SWE-Bench Pro	62.5 · score	Claude Code · Qwen refined and corrected task set	Source
SWE-Bench Multilingual	81 · score	mini-swe-agent	Source
NL2Repo	48.1 · score	Claude Code	Source
CoWorkBench	73.9 · score	internal	Source
JobBench	55.7 · score	—	Source
Agents' Last Exam	24.3 · pass@1	—	Source
Agents' Last Exam	51.2 · score	—	Source
Toolathlon-Verified	73.5 · pass@1	—	Source
IFBench	81.3 · score	—	Source
GPQA Diamond	91.7 · score	—	Source
Humanity's Last Exam	35.9 · score	without tools; GPT-4o judge	Source
LiveCodeBench	91.9 · score	6	Source
ClawEval-MM	64.4 · pass@3	—	Source
ClawEval-MM	60.4 · average score	—	Source
RecreationBench	49.9 · score	internal	Source
AndroidWorld	84.5 · score	—	Source
OSWorld	19.4 · binary completion rate	2.0	Source
OSWorld	52.3 · partial score	2.0	Source
Vision2Web	64 · score	Claude Code · gpt-5.4-2026-03-05 judge	Source
ERQA	72.3 · score	—	Source
LVBench	76.6 · score	—	Source
RealWorldQA	88.5 · score	—	Source
MathVision	90.6 · score	corrected annotations · without code interpreter	Source
MathVision	95.7 · score	corrected annotations · with code interpreter	Source
CharXiv Reasoning	84.6 · score	without code interpreter	Source
CharXiv Reasoning	90.6 · score	with code interpreter	Source

Complete source record (JSON)

{
  "id": "alibaba/qwen3.8-flash-next",
  "name": "Qwen3.8 Flash Next",
  "description": "Open-weight experimental preview of the Qwen4 architecture: hybrid-attention MoE (125B total, 6B active) with vision encoder for coding, agent tasks, and image and video understanding",
  "family": "qwen",
  "last_updated": "2026-08-27",
  "attachment": true,
  "reasoning": true,
  "tool_call": true,
  "structured_output": true,
  "temperature": true,
  "modalities": {
    "input": [
      "text",
      "image",
      "video"
    ],
    "output": [
      "text"
    ]
  },
  "limit": {
    "context": 262144,
    "output": 131072
  },
  "benchmarks": [
    {
      "name": "DeepSWE",
      "score": 58.7,
      "source": "https://huggingface.co/Qwen/Qwen3.8-Flash-Next",
      "metric": "score",
      "harness": "mini-swe-agent",
      "version": "1.1"
    },
    {
      "name": "SWE-Bench Pro",
      "score": 62.5,
      "source": "https://huggingface.co/Qwen/Qwen3.8-Flash-Next",
      "metric": "score",
      "harness": "Claude Code",
      "dataset": "Qwen refined and corrected task set"
    },
    {
      "name": "SWE-Bench Multilingual",
      "score": 81,
      "source": "https://huggingface.co/Qwen/Qwen3.8-Flash-Next",
      "metric": "score",
      "harness": "mini-swe-agent"
    },
    {
      "name": "NL2Repo",
      "score": 48.1,
      "source": "https://huggingface.co/Qwen/Qwen3.8-Flash-Next",
      "metric": "score",
      "harness": "Claude Code"
    },
    {
      "name": "CoWorkBench",
      "score": 73.9,
      "source": "https://huggingface.co/Qwen/Qwen3.8-Flash-Next",
      "metric": "score",
      "dataset": "internal"
    },
    {
      "name": "JobBench",
      "score": 55.7,
      "source": "https://huggingface.co/Qwen/Qwen3.8-Flash-Next",
      "metric": "score"
    },
    {
      "name": "Agents' Last Exam",
      "score": 24.3,
      "source": "https://huggingface.co/Qwen/Qwen3.8-Flash-Next",
      "metric": "pass@1"
    },
    {
      "name": "Agents' Last Exam",
      "score": 51.2,
      "source": "https://huggingface.co/Qwen/Qwen3.8-Flash-Next",
      "metric": "score"
    },
    {
      "name": "Toolathlon-Verified",
      "score": 73.5,
      "source": "https://huggingface.co/Qwen/Qwen3.8-Flash-Next",
      "metric": "pass@1"
    },
    {
      "name": "IFBench",
      "score": 81.3,
      "source": "https://huggingface.co/Qwen/Qwen3.8-Flash-Next",
      "metric": "score"
    },
    {
      "name": "GPQA Diamond",
      "score": 91.7,
      "source": "https://huggingface.co/Qwen/Qwen3.8-Flash-Next",
      "metric": "score"
    },
    {
      "name": "Humanity's Last Exam",
      "score": 35.9,
      "source": "https://huggingface.co/Qwen/Qwen3.8-Flash-Next",
      "metric": "score",
      "variant": "without tools; GPT-4o judge"
    },
    {
      "name": "LiveCodeBench",
      "score": 91.9,
      "source": "https://huggingface.co/Qwen/Qwen3.8-Flash-Next",
      "metric": "score",
      "version": "6"
    },
    {
      "name": "ClawEval-MM",
      "score": 64.4,
      "source": "https://huggingface.co/Qwen/Qwen3.8-Flash-Next",
      "metric": "pass@3"
    },
    {
      "name": "ClawEval-MM",
      "score": 60.4,
      "source": "https://huggingface.co/Qwen/Qwen3.8-Flash-Next",
      "metric": "average score"
    },
    {
      "name": "RecreationBench",
      "score": 49.9,
      "source": "https://huggingface.co/Qwen/Qwen3.8-Flash-Next",
      "metric": "score",
      "dataset": "internal"
    },
    {
      "name": "AndroidWorld",
      "score": 84.5,
      "source": "https://huggingface.co/Qwen/Qwen3.8-Flash-Next",
      "metric": "score"
    },
    {
      "name": "OSWorld",
      "score": 19.4,
      "source": "https://huggingface.co/Qwen/Qwen3.8-Flash-Next",
      "metric": "binary completion rate",
      "version": "2.0"
    },
    {
      "name": "OSWorld",
      "score": 52.3,
      "source": "https://huggingface.co/Qwen/Qwen3.8-Flash-Next",
      "metric": "partial score",
      "version": "2.0"
    },
    {
      "name": "Vision2Web",
      "score": 64,
      "source": "https://huggingface.co/Qwen/Qwen3.8-Flash-Next",
      "metric": "score",
      "harness": "Claude Code",
      "variant": "gpt-5.4-2026-03-05 judge"
    },
    {
      "name": "ERQA",
      "score": 72.3,
      "source": "https://huggingface.co/Qwen/Qwen3.8-Flash-Next",
      "metric": "score"
    },
    {
      "name": "LVBench",
      "score": 76.6,
      "source": "https://huggingface.co/Qwen/Qwen3.8-Flash-Next",
      "metric": "score"
    },
    {
      "name": "RealWorldQA",
      "score": 88.5,
      "source": "https://huggingface.co/Qwen/Qwen3.8-Flash-Next",
      "metric": "score"
    },
    {
      "name": "MathVision",
      "score": 90.6,
      "source": "https://huggingface.co/Qwen/Qwen3.8-Flash-Next",
      "metric": "score",
      "variant": "without code interpreter",
      "dataset": "corrected annotations"
    },
    {
      "name": "MathVision",
      "score": 95.7,
      "source": "https://huggingface.co/Qwen/Qwen3.8-Flash-Next",
      "metric": "score",
      "variant": "with code interpreter",
      "dataset": "corrected annotations"
    },
    {
      "name": "CharXiv Reasoning",
      "score": 84.6,
      "source": "https://huggingface.co/Qwen/Qwen3.8-Flash-Next",
      "metric": "score",
      "variant": "without code interpreter"
    },
    {
      "name": "CharXiv Reasoning",
      "score": 90.6,
      "source": "https://huggingface.co/Qwen/Qwen3.8-Flash-Next",
      "metric": "score",
      "variant": "with code interpreter"
    }
  ],
  "release_date": "2026-08-27",
  "open_weights": true,
  "license": "qwen-community-1.0",
  "weights": [
    {
      "label": "Hugging Face",
      "url": "https://huggingface.co/Qwen/Qwen3.8-Flash-Next"
    }
  ]
}

Infrastructure

Provider availability

All providers

Provider endpoint data will appear after this model's first OpenRouter endpoint sync.

Deployment graph

Can I run this?

Vertical Tabs only shows compatibility when it has structured evidence tying this exact model to hardware, runtime, quantization, memory use, and observed performance.

Browse hardware