Over recent months, we have intensified our focus on developing foundation models that deliver exceptional utility and performance. Rax 4.5 represents a significant leap forward, integrating breakthroughs in multimodal learning, architectural efficiency, reinforcement learning scale, and global accessibility to empower developers and enterprises with unprecedented capability and efficiency. Rax 4.5 features the following enhancement: For more details, please refer to our blog post Rax 4.5. WMT24++: a harder subset of WMT24 after difficulty labeling and rebalancing; we report the averaged scores on 55 languages using XCOMET-XXL. Empty cells (--) indicate scores not yet available or not…
Qwen3.5-2B is an open-weight model for image and text to text from Qwen, released under Apache License 2.0. It has 2.3B parameters and a 262,144-token context. At 16-bit it needs about 5.5 GB of GPU memory, which fits on 1x MI300X from $1.85 an hour, at the lowest prices in the SAVRN Index. It draws 2.8M downloads a month.
Over recent months, we have intensified our focus on developing foundation models that deliver exceptional utility and performance.
Parameters2.3B
Context262,144
Weights4.5 GB
Licenseapache-2.0
AccessOpen weights
Monthly Downloads2.8M
Runs On
What it takes to serve Qwen3.5-2B (2.3B parameters): the memory its weights need at each precision, and the cheapest way to rent enough data-center GPUs to hold them.
| Precision | Weights | Memory needed | Cheapest setup | Per hour | Also fits |
|---|---|---|---|---|---|
| 16-bit | 4.5 GB | 5.5 GB | 1x MI300X (192 GB) Vultr |
$1.85 | 1x H100 $1.99 · 1x MI325X $2.00 |
| 8-bit | 2.3 GB | 2.7 GB | 1x MI300X (192 GB) Vultr |
$1.85 | 1x H100 $1.99 · 1x MI325X $2.00 |
| 4-bit | 1.1 GB | 1.4 GB | 1x MI300X (192 GB) Vultr |
$1.85 | 1x H100 $1.99 · 1x MI325X $2.00 |
Memory is the weights at that precision plus 20% for the runtime and a short context; a long context needs more. Prices are the lowest on-demand hourly rates in the SAVRN Index, read Oct 16, 2026.
Qwen3.5-2B on every accelerator the SAVRN Index prices, at every precision
SAVRN's Notes on Qwen3.5-2B
2.3B parameters, image and text in, text out, over a 262,144-token context: 5.5 GB at 16-bit, 2.7 GB at 8-bit and 1.4 GB at 4-bit on the one 192 GB MI300X at $1.85 an hour that anchors our small-model pricing — the weights take a sliver and the long context fills the rest. This is the entry point of Qwen's natively multimodal 3.5 line, first published 2026-02-28 at 4.8M downloads a month.
Apache 2.0 covers commercial use, modification and redistribution. Reported scores — GPQA Diamond 44.9495, MMLU-Pro 55.3 — set honest expectations: this is the edge and appliance tier, and the derived list includes Rockchip RK182X LLM and VLM ports, which tells you where the community is taking it. For a bigger multimodal step, the 4B sibling doubles capacity for roughly twice the memory.
Model Card
By Qwen, published under apache-2.0, revision 15852e8c1636.
[!Note] This repository contains model weights and configuration files for the post-trained model in the Hugging Face Transformers format.
These artifacts are compatible with Hugging Face Transformers, vLLM, SGLang, KTransformers, etc.
In light of its parameter scale, the intended use cases are prototyping, task-specific fine-tuning, and other research or development purposes.
Over recent months, we have intensified our focus on developing foundation models that deliver exceptional utility and performance. Qwen3.5 represents a significant leap forward, integrating breakthroughs in multimodal learning, architectural efficiency, reinforcement learning scale, and global accessibility to empower developers and enterprises with unprecedented capability and efficiency.
Qwen3.5 Highlights
Qwen3.5 features the following enhancement:
Configuration
- Architecture
- Qwen3_5ForConditionalGeneration
- Context length (tokens)
- 262,144
- Layers
- 24
- Hidden size
- 2,048
- Feed-forward size
- 6,144
- Attention heads
- 8
- Key/value heads
- 2
- Head dimension
- 256
- Vocabulary size
- 248,320
- Model type
- qwen3_5
Identity and Version
- Repository
- Qwen/Qwen3.5-2B
- Publisher
- Qwen
- Task
- Image and text to text
- Modality
- Image and text
- Library
- transformers
- Parameters
- 2.3B parameters
- Languages
- Not stated by the source
- Revision
- 15852e8c16360a2fea060d615a32b45270f8a8fc
- First published
- 2026-02-28
- Last updated
- 2026-03-02
Files and Weights
13 files, 4.6 GB in total. The weights are 1 file totalling 4.5 GB in safetensors.
Weights1 file · 4.5 GB
Configuration4 files · 68.1 KB
Tokenizer4 files · 22.9 MB
Documentation2 files · 74.4 KB
Other1 file · 7.8 KB
Repository1 file · 1.6 KB
Every file
| File | Type | Size | SHA-256 |
|---|---|---|---|
| model.safetensors-00001-of-00001.safetensors | Weights | 4.5 GB | aa33250c4fc6 |
| config.json | Configuration | 2.9 KB | — |
| model.safetensors.index.json | Configuration | 64.5 KB | — |
| preprocessor_config.json | Configuration | 390 B | — |
| video_preprocessor_config.json | Configuration | 385 B | — |
| LICENSE | Documentation | 11.5 KB | — |
| README.md | Documentation | 62.8 KB | — |
| chat_template.jinja | Other | 7.8 KB | — |
| .gitattributes | Repository | 1.6 KB | — |
| merges.txt | Tokenizer | 3.4 MB | — |
| tokenizer.json | Tokenizer | 12.8 MB | 5f9e4d4901a9 |
| tokenizer_config.json | Tokenizer | 16.7 KB | — |
| vocab.json | Tokenizer | 6.7 MB | — |
License and Download
- License
- apache-2.0
- Access
- Open weights, no gate
- Download size
- 4.5 GB
Download from Qwen
Released by Qwen through ModelScope. Read the license.
Built From
- Derived from Qwen/Qwen3.5-2B-Base
Evaluations
Each result is shown as reported, with the conditions its reporter stated. None is a SAVRN measurement. A comparison lines two results up only when their configuration, unit and setup are all stated and identical.
| Benchmark | Conditions | Result | Reported by | Revision | Date |
|---|---|---|---|---|---|
| Idavidrein/gpqa | Task diamondMetric diamondSetup GPQA DiamondComparison conditions not established | 44.9495 | EvalEval Reported by a third party |
Evaluated revision not stated | 2026-04-20 |
| LiquidAI/ifstruct-v1.0 | Task ifstruct_v1Metric ifstruct_v1Comparison conditions not established | 33.15 | Liquid AI — IFStruct v1.0 blog (Qwen3.5-2B) Reported by a third party |
Evaluated revision not stated | 2026-06-30 |
| MMMU/MMMU_Pro | Task mmmu_pro_visionMetric mmmu_pro_visionSetup ThinkingComparison conditions not established | 50.3 | Model Card Reported by a third party |
Evaluated revision not stated | 2026-04-28 |
| TIGER-Lab/MMLU-Pro | Task mmlu_proMetric mmlu_proComparison conditions not established | 55.3 | Model Card Reported by a third party |
Evaluated revision not stated | 2026-03-02 |
| likaixin/ScreenSpot-Pro | Task overallMetric overallComparison conditions not established | 54.5 | Model Card Reported by a third party |
Evaluated revision not stated | 2026-03-18 |
| llamaindex/ExtractBench | Task longMetric longSetup Pipeline name: qwen3_5_2b_vllm_extract_oneshot_structured_output_fileComparison conditions not established | 25.93 | ExtractBench Reported by a third party |
Evaluated revision not stated | 2026-08-24 |
| llamaindex/ExtractBench | Task meanMetric meanSetup Pipeline name: qwen3_5_2b_vllm_extract_oneshot_structured_output_fileComparison conditions not established | 63.4 | ExtractBench Reported by a third party |
Evaluated revision not stated | 2026-08-24 |
| llamaindex/ExtractBench | Task mediumMetric mediumSetup Pipeline name: qwen3_5_2b_vllm_extract_oneshot_structured_output_fileComparison conditions not established | 50.93 | ExtractBench Reported by a third party |
Evaluated revision not stated | 2026-08-24 |
| llamaindex/ExtractBench | Task shortMetric shortSetup Pipeline name: qwen3_5_2b_vllm_extract_oneshot_structured_output_fileComparison conditions not established | 71.22 | ExtractBench Reported by a third party |
Evaluated revision not stated | 2026-08-24 |
| llamaindex/ParseBench | Task chartMetric chartSetup Pipeline name: qwen3_5_2b_vllm_layoutComparison conditions not established | 0.1 | ParseBench Reported by a third party |
Evaluated revision not stated | 2026-04-22 |
| llamaindex/ParseBench | Task layoutMetric layoutSetup Pipeline name: qwen3_5_2b_vllm_layoutComparison conditions not established | 18.3 | ParseBench Reported by a third party |
Evaluated revision not stated | 2026-04-22 |
| llamaindex/ParseBench | Task meanMetric meanSetup Pipeline name: qwen3_5_2b_vllm_layoutComparison conditions not established | 27.3 | ParseBench Reported by a third party |
Evaluated revision not stated | 2026-04-22 |
| llamaindex/ParseBench | Task tableMetric tableSetup Pipeline name: qwen3_5_2b_vllm_layoutComparison conditions not established | 0 | ParseBench Reported by a third party |
Evaluated revision not stated | 2026-04-22 |
| llamaindex/ParseBench | Task text_contentMetric text_contentSetup Pipeline name: qwen3_5_2b_vllm_layoutComparison conditions not established | 87.2 | ParseBench Reported by a third party |
Evaluated revision not stated | 2026-04-22 |
| llamaindex/ParseBench | Task text_formattingMetric text_formattingSetup Pipeline name: qwen3_5_2b_vllm_layoutComparison conditions not established | 31.1 | ParseBench Reported by a third party |
Evaluated revision not stated | 2026-04-22 |
Memory Requirements
| Precision | Weights in memory |
|---|---|
| As published | 4.5 GB |
| 16-bit | 4.5 GB |
| 8-bit | 2.3 GB |
| 4-bit | 1.1 GB |
Weights only, from the published parameter count; the key-value cache and runtime add to this.
Built on This Model
- Derived fromdQwen3.5-2B-Base
- Derived fromValen-2B
- Derived fromQwen3.5-2B-DSpark
- Derived fromDecision-1.0-Sol-2B
- Quantized fromRK182X-LLM-Qwen3.5-2B
- Derived fromRK182X-LLM-Qwen3.5-2B
- Derived fromQwen3.5-2B-auto-optimized
- Adapter ofdual-loop-qwen3.5-2b
- Derived fromdual-loop-qwen3.5-2b
- Quantized fromwag-2b
- Derived fromwag-2b
- Adapter ofweather-rescue-brazil-reader
- Derived fromweather-rescue-brazil-reader
- Adapter ofvnpen-writer-2b-v0.1-preview
- Derived fromvnpen-writer-2b-v0.1-preview
- Quantized fromRK182X-VLM-Qwen3.5-2B
- Derived fromRK182X-VLM-Qwen3.5-2B
- Derived fromDN-MOPD-Qwen3.5-2B-baseline-label-160updates
- Derived fromDN-MOPD-Qwen3.5-2B-160updates
- Derived fromThinkLess-2B
- Derived fromQwen3.5-2B-MLP
- Derived fromMaLoW-Qwen3.5-2B-MLP
- Derived fromMaLoW-Qwen3.5-2B
- Derived fromQwen3.5-2B-tool-router-v2-q4f16_1-MLC
- Derived fromQwen3.5-2B-tool-router-v2-q4f32_1-MLC
- Adapter ofudm_qwen3.5-2b_v2
- Derived fromudm_qwen3.5-2b_v2
- Derived fromuratori-ja-2b
- Derived fromOpenSLA
- Quantized fromQwen3.5-2B-RCOL
- Derived fromQwen3.5-2B-RCOL
- Quantized fromQwen3.5-2B-JNS
- Derived fromQwen3.5-2B-JNS
- Derived fromd3-nano
Compare Qwen3.5-2B
Questions About Qwen3.5-2B
How much GPU memory does Qwen3.5-2B need?
About 5.5 GB at 16-bit and 1.4 GB at 4-bit: the weights (2.3B parameters) plus a working margin. A long context needs more.
What is the cheapest GPU to run Qwen3.5-2B on?
At 16-bit, 1x MI300X from $1.85 an hour; at 4-bit, 1x MI300X from $1.85 an hour, at the lowest on-demand prices the SAVRN Index lists.
Can I use Qwen3.5-2B commercially?
Yes. Qwen3.5-2B is released under Apache License 2.0. The Apache License 2.0 is a permissive open-source license. It permits commercial use, modification and redistribution. It requires keeping the license and copyright notices and any NOTICE file, stating significant changes, and it includes an express patent grant from contributors.
What is Qwen3.5-2B's context length?
262,144 tokens, from the maximum position embeddings in its published configuration.
Similar Models
This is Qwen3.5-2B auto-optimized by Claude Fable for fast single-request text generation. Fable built and tuned the included qwen35fast inference engine while keeping Qwen's original BF16 weights unchanged. Across 12 development workloads, the Fable engine delivered 14× the decode speed of Transformers eager and 1.02× the speed of vLLM with MTP (geometric means). On 12 held-out workloads, it reached 528–866 tokens/s and 1.01× vLLM with MTP. Use Python 3.12 and an NVIDIA CUDA GPU. Download the model and install its dependencies: The original checkpoint also works with Transformers for Qwen's standard text and vision-language workflows; the speed figures above use qwen35fast. - Captured the…
Qwen3.5 with a two-layer MLP-Mixer decision head. Supports text, images and videos, and returns Choice, Noul and Score decisions. Multiple questions can share one state encoding with execution="sharedstate". Trained on millions of samples. Install Python 3.10+, PyTorch, Transformers 5.4.0, torchvision, Pillow and av. Flash Attention 2 is optional with a compatible CUDA build. The model contains its tokenizer, processor and custom inference code; a separate base-model download is unnecessary. Use attnimplementation="flashattention2" for Flash Attention, or execution="question" for independent questions. Video defaults to 16 frames. Image/video request examples and training instructions are…
Over recent months, we have intensified our focus on developing foundation models that deliver exceptional utility and performance. Qwen3.5 represents a significant leap forward, integrating breakthroughs in multimodal learning, architectural efficiency, reinforcement learning scale, and global accessibility to empower developers and enterprises with unprecedented capability and efficiency. For more details, please refer to our blog post Qwen3.5. If you find our work helpful, feel free to give us a cite.
This is the model card of a transformers model that has been pushed on the Hub. Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. Use the code below to get started with the model. Carbon emissions can be estimated using the Machine Learning Impact calculator presented in Lacoste et al. (2019).
The Label baseline of the DN-MOPD paper at Qwen3.5-2B continued to 160 updates (paper Table 5): multi-teacher on-policy distillation with label routing (each prompt is scored by the expert of its domain, every domain multiplier is 1). Released for comparison with DN-MOPD-Qwen3.5-2B; it is not the proposed method. Paper: Beyond Teacher Assignment: Domain-Normalized Multi-Teacher On-Policy Distillation (arXiv:2609.35347, project page) · Code: github.com/LiXin97/DN-MOPD The full recipe, with the launch scripts for every row of the paper's tables, is in recipes/qwen3.5/ and docs/recipe.md. This model was trained and evaluated with the non-thinking chat format. Pass enablethinking=False to the…