Qwen3 is the latest generation of large language models in Qwen series, offering a comprehensive suite of dense and mixture-of-experts (MoE) models. Built upon extensive training, Qwen3 delivers groundbreaking advancements in reasoning, instruction-following, agent capabilities, and multilingual support, with the following key features: - Uniquely support of seamless switching between thinking mode (for complex logical reasoning, math, and coding) and non-thinking mode (for efficient, general-purpose dialogue) within single model, ensuring optimal performance across various scenarios. - Significantly enhancement in its reasoning capabilities, surpassing previous QwQ (in thinking mode) and…
We introduce the updated version of the Qwen3-4B non-thinking mode, named Qwen3-4B-Instruct-2507, featuring the following key enhancements: - Significant improvements in general capabilities, including instruction following, logical reasoning, text…
Parameters4B
Context262,144
Weights8.0 GB
Licenseapache-2.0
AccessOpen weights
Monthly Downloads4M
Runs On
What it takes to serve Qwen3-4B-Instruct-2507 (4B 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 | 8.0 GB | 9.7 GB | 1x MI300X (192 GB) Vultr |
$1.85 | 1x H100 $1.99 · 1x MI325X $2.00 |
| 8-bit | 4.0 GB | 4.8 GB | 1x MI300X (192 GB) Vultr |
$1.85 | 1x H100 $1.99 · 1x MI325X $2.00 |
| 4-bit | 2.0 GB | 2.4 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 Sep 18, 2026.
Model Card
By Qwen, published under apache-2.0, revision cdbee75f17c0.
Highlights
We introduce the updated version of the Qwen3-4B non-thinking mode, named Qwen3-4B-Instruct-2507, featuring the following key enhancements:
- Significant improvements in general capabilities, including instruction following, logical reasoning, text comprehension, mathematics, science, coding and tool usage.
- Substantial gains in long-tail knowledge coverage across multiple languages.
- Markedly better alignment with user preferences in subjective and open-ended tasks, enabling more helpful responses and higher-quality text generation.
- Enhanced capabilities in 256K long-context understanding.
Model Overview
Qwen3-4B-Instruct-2507 has the following features: - Type: Causal Language Models - Training Stage: Pretraining & Post-training - Number of Parameters: 4.0B - Number of Paramaters (Non-Embedding): 3.6B - Number of Layers: 36 - Number of Attention Heads (GQA): 32 for Q and 8 for KV - Context Length: 262,144 natively.
NOTE: This model supports only non-thinking mode and does not generate <think></think> blocks in its output. Meanwhile, specifying enable_thinking=False is no longer required.
Configuration
- Architecture
- Qwen3ForCausalLM
- Context length (tokens)
- 262,144
- Layers
- 36
- Hidden size
- 2,560
- Feed-forward size
- 9,728
- Attention heads
- 32
- Key/value heads
- 8
- Head dimension
- 128
- Vocabulary size
- 151,936
- RoPE base
- 5,000,000
- Stored precision
- bfloat16
- Model type
- qwen3
Identity and Version
- Repository
- Qwen/Qwen3-4B-Instruct-2507
- Publisher
- Qwen
- Task
- Text generation
- Modality
- Text
- Library
- transformers
- Parameters
- 4B parameters
- Languages
- Not stated by the source
- Revision
- cdbee75f17c01a7cc42f958dc650907174af0554
- First published
- 2025-08-05
- Last updated
- 2025-09-17
Files and Weights
13 files, 8.1 GB in total. The weights are 3 files totalling 8.0 GB in safetensors.
Weights3 files · 8.0 GB
Configuration3 files · 33.8 KB
Tokenizer4 files · 15.9 MB
Documentation2 files · 19.5 KB
Repository1 file · 1.6 KB
Every file
| File | Type | Size | SHA-256 |
|---|---|---|---|
| model-00001-of-00003.safetensors | Weights | 4.0 GB | 75311d91bb08 |
| model-00002-of-00003.safetensors | Weights | 4.0 GB | 0b48adbb1f60 |
| model-00003-of-00003.safetensors | Weights | 99.6 MB | 7dd39ccca5e4 |
| config.json | Configuration | 727 B | — |
| generation_config.json | Configuration | 238 B | — |
| model.safetensors.index.json | Configuration | 32.8 KB | — |
| LICENSE | Documentation | 11.3 KB | — |
| README.md | Documentation | 8.2 KB | — |
| .gitattributes | Repository | 1.6 KB | — |
| merges.txt | Tokenizer | 1.7 MB | — |
| tokenizer.json | Tokenizer | 11.4 MB | aeb13307a71a |
| tokenizer_config.json | Tokenizer | 9.4 KB | — |
| vocab.json | Tokenizer | 2.8 MB | — |
License and Download
- License
- apache-2.0
- Access
- Open weights, no gate
- Download size
- 8.0 GB
Download from Qwen
Released by Qwen through ModelScope. Read the license.
Built From
- Described by arXiv:2505.09388
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 diamondComparison conditions not established | 62 | Model Card Reported by a third party |
Evaluated revision not stated | 2026-01-27 |
| TIGER-Lab/MMLU-Pro | Task mmlu_proMetric mmlu_proComparison conditions not established | 69.6 | Model Card Reported by a third party |
Evaluated revision not stated | 2026-02-03 |
Memory Requirements
| Precision | Weights in memory |
|---|---|
| As published | 8.0 GB |
| 16-bit | 8.0 GB |
| 8-bit | 4.0 GB |
| 4-bit | 2.0 GB |
Weights only, from the published parameter count; the key-value cache and runtime add to this.
Hosted Prices
| Host | Input / output | Unit | Observed |
|---|---|---|---|
| Nscale | $0.01 / $0.03 | input / output, per million tokens | Sep 18, 2026 |
From the SAVRN Index.
Built on This Model
- Quantized fromQwen3-4B-Instruct-2507-FP8
- Derived fromQwen3-4B-Instruct-2507-FP8
- Derived fromcobalt-seeded-rl-base-ramp25-stoppen-gen4k-ep2-ncp10-q4v3-groot16
- Derived fromcobalt-seeded-rl-base-ramp25-stoppen-gen4k-ep2-ncp10-base-q4v3
- Adapter ofstride-qwen3-4b-stabilized-2048-local_positive-alpha2-20260916
- Derived fromstride-qwen3-4b-stabilized-2048-local_positive-alpha2-20260916
- Adapter ofstride-qwen3-4b-stabilized-2048-local_positive-20260916
- Derived fromstride-qwen3-4b-stabilized-2048-local_positive-20260916
- Adapter ofstride-qwen3-4b-stabilized-2048-local_positive-alpha3-20260917
- Derived fromstride-qwen3-4b-stabilized-2048-local_positive-alpha3-20260917
- Adapter ofstride-qwen3-4b-stabilized-2048-token_uniform-20260916
- Derived fromstride-qwen3-4b-stabilized-2048-token_uniform-20260916
- Adapter ofstride-qwen3-4b-stabilized-2048-single_step-20260916
- Derived fromstride-qwen3-4b-stabilized-2048-single_step-20260916
- Adapter ofstride-qwen3-4b-stabilized-2048-local_positive-alpha05-20260917
- Derived fromstride-qwen3-4b-stabilized-2048-local_positive-alpha05-20260917
- Adapter ofstride-qwen3-4b-stabilized-2048-grpo-20260916
- Derived fromstride-qwen3-4b-stabilized-2048-grpo-20260916
- Derived fromquell-v2
Compare Qwen3-4B-Instruct-2507
Questions About Qwen3-4B-Instruct-2507
How much GPU memory does Qwen3-4B-Instruct-2507 need?
About 9.7 GB at 16-bit and 2.4 GB at 4-bit: the weights (4B parameters) plus a working margin. A long context needs more.
What is the cheapest GPU to run Qwen3-4B-Instruct-2507 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 Qwen3-4B-Instruct-2507 commercially?
Yes. Qwen3-4B-Instruct-2507 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 Qwen3-4B-Instruct-2507's context length?
262,144 tokens, from the maximum position embeddings in its published configuration.
Similar Models
Qwen3 is the latest generation of large language models in Qwen series, offering a comprehensive suite of dense and mixture-of-experts (MoE) models. Building upon extensive advancements in training data, model architecture, and optimization techniques, Qwen3 delivers the following key improvements over the previously released Qwen2.5: Qwen3-4B-Base has the following features: For more details, including benchmark evaluation, hardware requirements, and inference performance, please refer to our blog, GitHub, and Documentation. The code of Qwen3 has been in the latest Hugging Face transformers and we advise you to use the latest version of transformers. With transformers<4.51.0, you will…
Qwen3 is the latest generation of large language models in Qwen series, offering a comprehensive suite of dense and mixture-of-experts (MoE) models. Building upon extensive advancements in training data, model architecture, and optimization techniques, Qwen3 delivers the following key improvements over the previously released Qwen2.5: Qwen3-4B-Base has the following features: For more details, including benchmark evaluation, hardware requirements, and inference performance, please refer to our blog, GitHub, and Documentation. The code of Qwen3 has been in the latest Hugging Face transformers and we advise you to use the latest version of transformers. With transformers<4.51.0, you will…
We introduce the updated version of the Qwen3-4B non-thinking mode, named Qwen3-4B-Instruct-2507, featuring the following key enhancements: - Significant improvements in general capabilities, including instruction following, logical reasoning, text comprehension, mathematics, science, coding and tool usage. - Substantial gains in long-tail knowledge coverage across multiple languages. - Markedly better alignment with user preferences in subjective and open-ended tasks, enabling more helpful responses and higher-quality text generation. - Enhanced capabilities in 256K long-context understanding. Qwen3-4B-Instruct-2507 has the following features: NOTE: This model supports only non-thinking…
This model is a fine-tuned version of cosmos1030/gmp-kd3e-1-s80pct-lr1e-420260916220740 on the trl-lib/ultrafeedbackbinarized dataset. It has been trained using TRL. This model was trained with DPO, a method introduced in Direct Preference Optimization: Your Language Model is Secretly a Reward Model.
The full ImmuneCoT method fuses the two safety branches with a Base-adjusted product-of-experts Qimm(v) ∝ q̃rec(v)·q̃res(v)/qB(v). This checkpoint uses the naive fusion Qno-base(v) ∝ q̃rec(v)·q̃res(v) — the same branch weights (λrec=0.5, λres=0.7) but no division by the base distribution — isolating whether the gains come from combining Recognition+Response at all, or specifically from the Base-adjusted PoE term. Intended use: research reproducibility for the ImmuneCoT paper's RQ3 ablation.