Qwen

Qwen3.6 35B A3B

Tool callingStructured outputReasoningImage inputAudio inputVideo input
Qwen3.6 35B A3B model card cover

Open multimodal Qwen MoE for local agents that need vision, audio, and code

Context
262K
Released
6mo ago
Parameters
35B
Input / 1M tokens
$0.15
Output / 1M tokens
$1.00
Cached input / 1M tokens
$0.05

Sources:models.devOpenRouter

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.

Other published benchmarks

EvaluationReported score and source
FrontierMath-Tiers-1-3-v2-Private
17.5%
Epoch AI · qwen3.6-35b-a3b
Epoch AI evaluation
FrontierMath-Tiers-1-3-v2-Private
20.4%
Epoch AI · qwen3.6-35b-a3b_none
Reasoning effort: none
Epoch AI evaluation
GPQA diamond
83.8%
Epoch AI · qwen3.6-35b-a3b
Epoch AI evaluation
GPQA diamond
84.8%
Epoch AI · qwen3.6-35b-a3b_none
Reasoning effort: none
Epoch AI evaluation
OTIS Mock AIME 2024-2025
86.7%
Epoch AI · qwen3.6-35b-a3b
Epoch AI evaluation
OTIS Mock AIME 2024-2025
68.9%
Epoch AI · qwen3.6-35b-a3b_none
Reasoning effort: none
Epoch AI evaluation
Terminal Bench
24.6%
https://www.tbench.ai/leaderboard/terminal-bench/2.0 · qwen3.6-35b-a3b
Agent: little-coder
https://www.tbench.ai/leaderboard/terminal-bench/2.0
Original-source terms apply

Benchmark compilation: Epoch AI, Capabilities & benchmarking (CC BY attribution). Scores are converted to percentages and model names matched to the Vertical Tabs catalog. Individual evaluations are credited to their original sources. External results may carry separate licences and evaluation configurations; do not interpret scores from different harnesses as directly comparable.

Specifications

Developer
Qwen
Model family
Qwen
Type
Language model
Predecessor
—
Successor
—
Released
April 17, 2026
Model ID
qwen/qwen3.6-35b-a3b
Openness
Open weights
License
—
Parameters
35B
Input modalities
text, image, video
Output modalities
text
Modality
text+image+video->text
Native context
256K
Tokenizer
Qwen
Max output tokens
236K
Source freshness
Metadata synced Oct 8, 2026, 10:17 PM UTC

Open-weight builds

Ollama variants

Every published Ollama tag matched to this model, including parameter editions, instruction/base variants, and quantized builds.

30 builds
VariantParametersTypeQuantizationSizeContextInput
qwen3.6:27b-bf1627B—BF1656 GB256Ktext + image
qwen3.6:27b-coding-bf1627B—BF1655 GB256Ktext + image
qwen3.6:27b-mlx-bf1627B—BF1656 GB256Ktext + image
qwen3.6:27b-mtp-bf1627B—BF1656 GB256Ktext + image
qwen3.6:27b-coding-mtp-q4_K_M27B—Q4_K_M18 GB256Ktext + image
qwen3.6:27b-mtp-q4_K_M27B—Q4_K_M18 GB256Ktext + image
qwen3.6:27b-q4_K_M27B—Q4_K_M17 GB256Ktext + image
qwen3.6:27b-mtp-q8_027B—Q8_030 GB256Ktext + image
qwen3.6:27b-q8_027B—Q8_030 GB256Ktext + image
qwen3.6:27b27B——18 GB256Ktext + image
qwen3.6:27b-coding27B——18 GB256Ktext + image
qwen3.6:27b-coding-mxfp827B——30 GB256Ktext + image
qwen3.6:27b-coding-nvfp427B——19 GB256Ktext + image
qwen3.6:27b-mlx27B——19 GB256Ktext + image
qwen3.6:27b-mxfp827B——30 GB256Ktext + image
qwen3.6:35b-a3b-bf1635B—BF1671 GB256Ktext + image
qwen3.6:35b-a3b-coding-bf1635B—BF1670 GB256Ktext + image
qwen3.6:35b-a3b-mlx-bf1635B—BF1672 GB256Ktext + image
qwen3.6:35b-a3b-mtp-bf1635B—BF1672 GB256Ktext + image
qwen3.6:35b-a3b-coding-mtp-q4_K_M35B—Q4_K_M23 GB256Ktext + image
qwen3.6:35b-a3b-mtp-q4_K_M35B—Q4_K_M23 GB256Ktext + image
qwen3.6:35b-a3b-q4_K_M35B—Q4_K_M24 GB256Ktext + image
qwen3.6:35b-a3b-mtp-q8_035B—Q8_039 GB256Ktext + image
qwen3.6:35b-a3b-q8_035B—Q8_039 GB256Ktext + image
qwen3.6:35b-a3b-coding35B——23 GB256Ktext + image
qwen3.6:35b-a3b-coding-mxfp835B——38 GB256Ktext + image
qwen3.6:35b-a3b-coding-nvfp435B——24 GB256Ktext + image
qwen3.6:35b-a3b-mxfp835B——38 GB256Ktext + image
qwen3.6:35b-a3b-nvfp435B——24 GB256Ktext + image
qwen3.6:latest35B——23 GB256Ktext + image
Additional source evidence
Source model
alibaba/qwen3.6-35b-a3b
Source family
qwen
Last updated
2026-04-17
Model context
262K
Model output limit
66K
Open weights
Yes
File attachments
Yes
Reasoning
Yes
Tool calling
Yes
Structured output
Yes
Temperature control
Yes

Downloadable weights

Published benchmarks

BenchmarkScoreEvaluation detailsEvidence
SWE-Bench Verified73.4 · resolved—Source
Complete source record (JSON)
{
  "id": "alibaba/qwen3.6-35b-a3b",
  "name": "Qwen3.6 35B-A3B",
  "description": "Open multimodal Qwen MoE for local agents that need vision, audio, and code",
  "family": "qwen",
  "last_updated": "2026-04-17",
  "attachment": true,
  "reasoning": true,
  "tool_call": true,
  "structured_output": true,
  "temperature": true,
  "modalities": {
    "input": [
      "text",
      "image",
      "video",
      "audio"
    ],
    "output": [
      "text"
    ]
  },
  "limit": {
    "context": 262144,
    "output": 65536
  },
  "benchmarks": [
    {
      "name": "SWE-Bench Verified",
      "score": 73.4,
      "source": "https://huggingface.co/Qwen/Qwen3.6-35B-A3B",
      "metric": "resolved"
    }
  ],
  "release_date": "2026-04-17",
  "open_weights": true,
  "weights": [
    {
      "label": "Hugging Face",
      "url": "https://huggingface.co/Qwen/Qwen3.6-35B-A3B"
    }
  ]
}

Infrastructure

Provider availability

All providers
ProviderEndpointContextInput /M tokensOutput /M tokensCached input /M tokensCache write /M tokensUptime 1dData handling
Darkbloomdarkbloom/fp4262K$0.05$0.70$0.02—97.36%
ZDR: No

Training: Unknown

Prompt logging: Unknown

Retention: Unknown

Checked 2026-10-15

AkashMLakashml/fp8262K$0.10$0.90$0.05—100.00%
ZDR: Yes

Training: Unknown

Prompt logging: Unknown

Retention: Unknown

Checked 2026-10-15

DeepInfradeepinfra/fp8262K$0.10$0.95$0.10—77.17%
ZDR: Yes

Training: Unknown

Prompt logging: Unknown

Retention: Unknown

Checked 2026-10-15

DekaLLMdekallm262K$0.10$1.00$0.05—97.94%
ZDR: Yes

Training: Unknown

Prompt logging: Unknown

Retention: Unknown

Checked 2026-10-15

Venicevenice/fp8256K$0.10$1.00——96.41%
ZDR: Yes

Training: Unknown

Prompt logging: Unknown

Retention: Unknown

Checked 2026-10-15

Parasailparasail/fp8262K$0.15$1.00$0.05—100.00%
ZDR: Yes

Training: Unknown

Prompt logging: Unknown

Retention: Unknown

Checked 2026-10-15

AtlasCloudatlas-cloud/fp8262K$0.19$1.11$0.19—85.32%
ZDR: No

Training: Unknown

Prompt logging: Unknown

Retention: Unknown

Checked 2026-10-15

Phalaphala262K$0.20$1.27$0.06—96.83%
ZDR: Yes

Training: Unknown

Prompt logging: Unknown

Retention: Unknown

Checked 2026-10-15

SiliconFlowsiliconflow/fp8262K$0.24$1.80$0.15—97.38%
ZDR: Yes

Training: Unknown

Prompt logging: Unknown

Retention: Unknown

Checked 2026-10-15

CoreWeavecoreweave/fp8262K$0.25$1.25$0.25—100.00%
ZDR: Yes

Training: Unknown

Prompt logging: Unknown

Retention: Unknown

Checked 2026-10-15

Provider data:OpenRouter

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