Qwen3 4B Instruct 2507
Alibaba · Qwen/Qwen3-4B-Instruct-2507
Q4 file 2.5 GB (Q4_K_M) · runs well from 8 GB · Weights license: Apache 2.0
Quality
No result in LiveBench 2026-06-25, LMArena 2026-10-22 Not in LiveBench 2026-06-25 or LMArena 2026-10-22
Reported by the publisher
- MMLU-Pro
- 69.6Knowledge
- MMLU-Redux
- 84.2Knowledge
- GPQA
- 62.0Knowledge
- SuperGPQA
- 42.8Knowledge
- AIME25
- 47.4Reasoning
- HMMT25
- 31.0Reasoning
- ZebraLogic
- 80.2Reasoning
- LiveBench 20241125
- 63.0Reasoning
Every reported result
- LiveCodeBench v6 (25.02-25.05)
- 35.1Coding
- MultiPL-E
- 76.8Coding
- Aider-Polyglot
- 12.9Coding
- IFEval
- 83.4Alignment
- Arena-Hard v2
- 43.4Alignment
- Creative Writing v3
- 83.5Alignment
- WritingBench
- 83.4Alignment
- BFCL-v3
- 61.9Agent
- TAU1-Retail
- 48.7Agent
- TAU1-Airline
- 32.0Agent
- TAU2-Retail
- 40.4Agent
- TAU2-Airline
- 24.0Agent
- TAU2-Telecom
- 13.2Agent
- MultiIF
- 69.0Multilingualism
- MMLU-ProX
- 61.6Multilingualism
- INCLUDE
- 60.1Multilingualism
- PolyMATH
- 31.1Multilingualism
Across Qwen3 sizes
| Benchmark (publisher settings) | 4B Instruct 2507 | 30B A3B Instruct 2507 |
|---|---|---|
| MMLU-Pro | 69.6 | 78.4 |
| MMLU-Redux | 84.2 | 89.3 |
| GPQA | 62.0 | 70.4 |
| SuperGPQA | 42.8 | 53.4 |
| AIME25 | 47.4 | 61.3 |
| HMMT25 | 31.0 | 43.0 |
| ZebraLogic | 80.2 | 90.0 |
| LiveBench 20241125 | 63.0 | 69.0 |
| LiveCodeBench v6 (25.02-25.05) | 35.1 | 43.2 |
| MultiPL-E | 76.8 | 83.8 |
| Aider-Polyglot | 12.9 | 35.6 |
| IFEval | 83.4 | 84.7 |
| Arena-Hard v2 | 43.4 | 69.0 |
| Creative Writing v3 | 83.5 | 86.0 |
| WritingBench | 83.4 | 85.5 |
| BFCL-v3 | 61.9 | 65.1 |
| TAU1-Retail | 48.7 | 59.1 |
| TAU1-Airline | 32.0 | 40.0 |
| TAU2-Retail | 40.4 | 57.0 |
| TAU2-Airline | 24.0 | 38.0 |
| TAU2-Telecom | 13.2 | 12.3 |
| MultiIF | 69.0 | 67.9 |
| MMLU-ProX | 61.6 | 72.0 |
| INCLUDE | 60.1 | 71.9 |
| PolyMATH | 31.1 | 43.1 |
Values as printed on the card, with the publisher's own settings (card section: Performance); column “Qwen3-4B-Instruct-2507”. Not comparable across publishers and not used in any ranking. Model card
Run it
Run it
File Q4_K_M · 2.5 GB · unsloth/Qwen3-4B-Instruct-2507-GGUF · What is Q4?
Measured 134.5 tok/s · MLX q4 · 1 machine · mlx-lm BENCHMARKS.md · all rows
ollama run hf.co/unsloth/Qwen3-4B-Instruct-2507-GGUF:Q4_K_M- Install Ollama
- Run the command above
- Ask:
Explain what a context window is, in two sentences.
Runs on
Fast memory holds it 118
The planning estimate fits the memory the model can use quickly: graphics card VRAM, or unified memory within the macOS GPU share.
Show all 118
Within memory, outside the fast part 1
Within the machine's memory, outside its fast part: on a graphics card the layers that do not fit run on the CPU from PC RAM (the listed RAM, or 32 GB assumed); on a Mac, above the macOS default GPU share. Not a speed figure.
Qwen3 4B Instruct 2507 is a text model from Alibaba with 4B parameters. It was released on August 6, 2025 and reads up to 262,144 tokens of context. Its Q4 weights plus working room come to about 3 GB before the machine's reserve; the smallest reference machines whose fast memory holds that are Mac mini M2 · 8 GB unified (~5 GB with its reserve) and RTX 4060 · 8 GB VRAM (~5 GB with its reserve). We collected 29 downloadable GGUF and MLX files from 3 publisher sources; the most downloaded GGUF source is unsloth's. It has no result in LiveBench 2026-06-25, so no quality score is shown.
Hardware profiles grouped by the Q4 planning estimate on each machine (dedicated GPUs reserve 2 GB; unified and system memory 25%, 2 to 8 GB): the fast memory holds it, it is within memory but outside the fast part (layers offloaded to PC RAM, or above the macOS GPU share), or the machine is one memory tier short. Each card opens its hardware profile. Not a measured fit or speed.
On the reference machines
| Where Q4 sits | ||||
|---|---|---|---|---|
| Mac mini M2 · 8 GB unifiedFast memory holds itQ6_K8,192 tokens | ~5 GB | Fast memory holds it | Q6_K | 8,192 tokens |
| Mac mini M4 · 16 GB unifiedFast memory holds itQ8_032,768 tokens | ~7 GB | Fast memory holds it | Q8_0 | 32,768 tokens |
| Mac mini M4 · 24 GB unifiedFast memory holds itQ8_065,536 tokens | ~9 GB | Fast memory holds it | Q8_0 | 65,536 tokens |
| Mac mini M4 · 32 GB unifiedFast memory holds itQ8_065,536 tokens | ~11 GB | Fast memory holds it | Q8_0 | 65,536 tokens |
| Mac mini M5 Pro · 64 GB unifiedFast memory holds itQ8_0262,144 tokens | ~11 GB | Fast memory holds it | Q8_0 | 262,144 tokens |
| Mac Studio M5 Max · 128 GB unifiedFast memory holds itQ8_0262,144 tokens | ~11 GB | Fast memory holds it | Q8_0 | 262,144 tokens |
| Mac Studio · M5 Ultra · 36-core CPU / 80-core GPU · 512 GB unifiedFast memory holds itQ8_0262,144 tokens | ~11 GB | Fast memory holds it | Q8_0 | 262,144 tokens |
| RTX 4060 · 8 GB VRAMFast memory holds itQ6_K8,192 tokens | ~5 GB | Fast memory holds it | Q6_K | 8,192 tokens |
| RTX 3060 · 12 GB VRAMFast memory holds itQ8_032,768 tokens | ~5 GB | Fast memory holds it | Q8_0 | 32,768 tokens |
| RTX 5080 · 16 GB VRAMFast memory holds itQ8_032,768 tokens | ~5 GB | Fast memory holds it | Q8_0 | 32,768 tokens |
| RTX 4090 · 24 GB VRAMFast memory holds itQ8_065,536 tokens | ~5 GB | Fast memory holds it | Q8_0 | 65,536 tokens |
| RTX 5090 · 32 GB VRAMFast memory holds itQ8_0131,072 tokens | ~5 GB | Fast memory holds it | Q8_0 | 131,072 tokens |
| DGX H100 · 8 × 80 GBFast memory holds itQ8_0262,144 tokens | ~19 GB | Fast memory holds it | Q8_0 | 262,144 tokens |
The file column uses the site's quant ladder (Q8_0 to Q3_K_M, MLX 8 to 4-bit on a Mac): the largest collected file whose size × 1.2 plus the context cache and the machine's reserve fits the fast memory. Planning estimates, not measured runs.
Compare Qwen3 4B Instruct 2507 using its publisher card. No overall score appears in the LiveBench 2026-06-25 snapshot.
The Q4 weights plus working room come to about 3 GB from 4B stated parameters, before the machine's reserve (Runs on adds it per machine). It is not a measured minimum; context, batch size, runtime and quantization change actual memory use.
Versions to download
Files you can download for this checkpoint. The section opens on the most downloaded collected GGUF source, unsloth, at Q4_K_M. On a Mac it opens on the most downloaded MLX source. Sizes are the Hub's published bytes. What is Q4?
unsloth
unsloth/Qwen3-4B-Instruct-2507-GGUF · Qwen3-4B-Instruct-2507-Q4_K_M.gguf · 2.5 GB
Run this file
ollama run hf.co/unsloth/Qwen3-4B-Instruct-2507-GGUF:Q4_K_Mllama-server -hf unsloth/Qwen3-4B-Instruct-2507-GGUF:Q4_K_Mlms get https://huggingface.co/unsloth/Qwen3-4B-Instruct-2507-GGUF- mlx-lm: mlx-lm runs MLX repositories; pick the MLX version on a Mac
Quality scores on this page belong to the original checkpoint; these files are not scored or ranked. File sizes read 2026-10-23 · download counts read 2026-10-17 · commands: Hugging Face Hub docs: Use Ollama with any GGUF model · llama.cpp server README (-hf <user>/<model>[:quant]) · LM Studio docs: lms get with a Hugging Face URL · mlx-lm README (mlx_lm.generate --model)
Fine-tunes, adapters and other versions
- Quantized versions
- 353
- Fine-tunes
- 2,615
- Adapters · merges
- Not counted yet: the Hub's base-model lists are read by the weekly collector.
Browse them on the Hugging Face model tree · read 2026-10-23
Uncensored and abliterated versions
Some community uploads remove the model's safety refusals. Quality may drop, the original publisher does not support them, claims about them are the uploader's, and the original license still applies. This site never recommends or ranks them.
22 derivative repositories have “uncensored” or “abliterated” in their name (read 2026-10-23).
Measured speed
Tokens per second by machine
Output tok/s · 1 machine · fastest 5 of 7 · every row in the table · What is a token?
| Apple M4 Max · 64 GB (GPU cores not stated) | 134.52 | MLX q4 | MLX (mlx-lm) | mlx-community/Qwen3-4B-Instruct-2507-4bit |
|---|---|---|---|---|
| Apple M4 Max · 64 GB (GPU cores not stated) | 126.00 | MLX q4 g32 | MLX (mlx-lm) | mlx-community/Qwen3-4B-Instruct-2507-4bit-g32 |
| Apple M4 Max · 64 GB (GPU cores not stated) | 116.39 | MLX q5 | MLX (mlx-lm) | mlx-community/Qwen3-4B-Instruct-2507-5bit |
| Apple M4 Max · 64 GB (GPU cores not stated) | 110.29 | MLX q5 g32 | MLX (mlx-lm) | mlx-community/Qwen3-4B-Instruct-2507-5bit-g32 |
| Apple M4 Max · 64 GB (GPU cores not stated) | 104.68 | MLX q6 | MLX (mlx-lm) | mlx-community/Qwen3-4B-Instruct-2507-6bit |
| Apple M4 Max · 64 GB (GPU cores not stated) | 86.907 | MLX q8 | MLX (mlx-lm) | mlx-community/Qwen3-4B-Instruct-2507-8bit |
| Apple M4 Max · 64 GB (GPU cores not stated) | 52.47 | MLX bf16 | MLX (mlx-lm) | Qwen/Qwen3-4B-Instruct-2507 |
Source: mlx-lm BENCHMARKS.md · maintainer result (checked 2026-10-25) · MLX (mlx-lm) · mlx 0.29.2.dev20251008+85a8824a8 · mlx-lm 0.28.2 · macOS 26.1 · not estimated
How to run
The memory needed above is the Q4 load before a machine's reserve. Formula: parameters × 0.625 bytes, plus 20% working allowance; each machine then adds its system reserve: 25% of unified or system memory, 2 to 8 GB, or 2 GB per GPU on dedicated cards. This is neither measured VRAM nor checkpoint download size.
Q4 planning estimate only. Check checkpoint format, runtime support, KV cache, context, batch size and offload before choosing hardware.
Evidence level: Generic estimate only; no execution reproduced by this site.
- Publisher sampling defaults
- temperature 0.7 · top_p 0.8 · top_k 20generation_config.json · read 2026-10-23source
- Chat template markers
- tool calls · no thinking markersmarkers found in the template, not a test of the feature · read 2026-10-23source
Before downloading
- Check the exact checkpoint and license linked here.
- Confirm a compatible runtime and precision or quantization for your platform.
- Allow for context, batch size and any extra media components beyond the weights.
Technical details
What the source and the catalog store for this checkpoint. Missing fields are left out, never zero.
- Checkpoint / version
- Qwen/Qwen3-4B-Instruct-2507
- Task and modality
- language and reasoning · text
- Count and basis
- 4B · manufacturer specificationQwen/Qwen3-4B-Instruct-2507 model card (Hugging Face) · read 2026-10-17source
- Context
- 262,144 tokens published for this modelQwen/Qwen3-4B-Instruct-2507 model card (Hugging Face) · read 2026-10-17source
- Architecture
- DenseQwen/Qwen3-4B-Instruct-2507 config.json (Hugging Face) · read 2026-10-17source
- Released
- QwenLM/Qwen3 README (GitHub) · read 2026-10-09source
- Runtime and precision
- No local execution verified.
- Commercial use
- Allowed Apache License 2.0 text (SPDX) · read 2026-10-18license text
Every stored field
- Revision checked
- cdbee75f17c01a7cc42f958dc650907174af0554
- Architecture
- Qwen3ForCausalLM
- Library
- transformers
- Hub task tag
- text-generation
- Distributions in this group
- mlx-community/Qwen3-4B-Instruct-2507-4bit
- Formats
- safetensors
- Access
- Open download
- Publisher kind
- Official
- Tensor evidence
- floating-tensor-elements · 4.02B · BF16checked 2026-09-26
- Repository created
- 2025-08-05creation is not the model's release
- Repository updated
- 2025-09-17a commit is not a release date
- Hub downloads (30 days) · likes
- 2,577,859 · 1,027this repository only · read 2026-10-25
- Catalog state
- Hub config or weights found; no inventory benchmark link
Sources
Identity and published specifications come from the checkpoint page. Benchmark scores belong to the named release and do not measure speed on your computer.
- Publisher checkpoint and model card
- Weight license and use terms
- Quality benchmark source · LiveBench 2026-06-25 overall
Catalog reviewed: 2026-10-17. Benchmark release: 2026-06-25; snapshot collected: 2026-10-18.






























































