Qwen3 8B — Hardware Requirements & GPU Compatibility
ChatFunctions
Qwen3 8B is an 8.2-billion parameter instruction-tuned model from Alibaba Cloud's Qwen 3 series. It is a general-purpose chat model that delivers strong performance across reasoning, multilingual understanding, and coding tasks while remaining efficient enough to run on consumer GPUs with 8GB or more of VRAM. Like other Qwen 3 models, it supports hybrid thinking mode for flexible reasoning depth. The model benefits from the improved pretraining data and training methodology of the Qwen 3 generation, offering notable quality gains over Qwen 2.5 at the same parameter count. It is widely supported by inference frameworks including llama.cpp, vLLM, and Ollama. Released under the Apache 2.0 license.
Runs great on a 24 GB GPU
Runs from 8 GB (Q4_K_M · 16K)Best on 24 GB (Q4_K_M · ~119 tok/s · 40K)Macs from 16 GB (Q4_K_M · ~25 tok/s)
Score 93.9#73 of 106 open#19 of 36 that fit 24 GBApache 2.0
Reference GPU · RTX 4090 24 GB
Estimates, ±20% · how we estimate
llmrun Composite Score →
93.9
- 90% range 90.5–97.3
- #73 of 106 open models
- #19 of 36 that fit 24 GB
Run Qwen3 8B
Best fit for your MacBook Air 15" M4 (16 GB). Every other method, OS and quant is in the step-by-step guide.Each way opens a step-by-step guide for your OS and quant.
Recommended for your MacBook Air 15" M4 (16 GB)
mlx_lm.chat --model mlx-community/Qwen3-8B-4bitMLX repo with its own quantisation
Download: Qwen3-8B-Q4_K_M.gguf5.03 GBQwen
Ways to runOther ways
each opens step-by-stepRun in cloud
Runs on RTX 3080 Ti (12 GB) (4 GB headroom) · Q4_K_M
- Generation speed
- ~107 tok/s
- generation speed
- Cost per 1M output tokens
- $0.32
- per 1M output tokens
How Much VRAM Does Qwen3 8B Need?
Select a quantization level; the listed hardware adheres to it.
| Quantization | VRAM | +32K | On your device | |
|---|---|---|---|---|
| 4.1 GB | 8.6 GB | 2-bit · K-quant — 2-bit quantization with K-quant improvements | Fits: 40K context · ~21 tok/s | |
| 4.2 GB | 8.7 GB | 3-bit small — 3-bit small quantization | Fits: 40K context · ~20 tok/s | |
| 4.6 GB | 9.1 GB | 3-bit medium — 3-bit medium quantization | Fits: 40K context · ~18 tok/s | |
| 4.7 GB | 9.2 GB | 4-bit · legacy — 4-bit legacy quantization | Fits: 40K context · ~18 tok/s | |
| 5.5 GB | 10.1 GB | 4-bit medium — 4-bit medium quantization — most popular sweet spot | Fits: 40K context · ~15 tok/s | |
| 6.4 GB | 11 GB | 5-bit medium — 5-bit medium quantization — good quality/size tradeoff | Fits: 40K context · ~13 tok/s | |
| 7.4 GB | 11.9 GB | 6-bit, very good — 6-bit quantization, very good quality | Fits: 32K context · ~11 tok/s | |
| 8.8 GB | 13.3 GB | 8-bit, near-lossless — 8-bit quantization, near-lossless | Fits: 16K context · ~10 tok/s |
Selected Q4_K_M
4-bit medium quantization7.6GBat 16K context
- Weights
- 4.9 GB
- KV cache
- 2.4 GB
- Overhead
- 0.3 GB
Fits MacBook Air 15" M4 (16 GB) · 4.9 GB free · ~15 tok/s
≥50 tok/s needs ≥424 GB/s · RTX 5090 ~211, RTX 3090 Ti ~119
Which Hardware Can Run Qwen3 8B?
Q4_K_M · 5.5 GBQwen3 8B (Q4_K_M) requires 5.5 GB of VRAM to load the model weights. For comfortable inference with headroom for KV cache and system overhead, 8+ GB is recommended. Using the full 41K context window can add up to 5.7 GB, bringing total usage to 11.3 GB. 52 GPUs can run it, including NVIDIA GeForce RTX 5090, NVIDIA GeForce RTX 3090 Ti, NVIDIA GeForce RTX 3070 Ti. 151 devices with unified memory can run Qwen3 8B, including Mac Studio (M3 Ultra, 256GB), Mac Studio (M3 Ultra, 512GB), NVIDIA Jetson Orin Nano 8GB (Super).
≤ 11 GB
≤ 11 GB tier →fits · 0.4 GB free · up to 16K ctx
16 GPUs · 1 device family
RTX 3080~90Simulate output on RTX 3080RTX 2080 Ti~73Simulate output on RTX 2080 TiRTX 3070 Ti~72Simulate output on RTX 3070 TiOrin Nano~12Simulate output on NVIDIA Jetson Orin Nano 8GB (Super)12 GB
12 GB tier →fits · 4.4 GB free · up to 40K ctx
10 GPUs · 1 device family
RTX 3080 Ti~107Simulate output on RTX 3080 TiRTX 5070~79Simulate output on RTX 5070RTX 4070~59Simulate output on RTX 4070iPhone~10Simulate output on Apple iPhone 17 Pro20 GB
fits · 12.4 GB free · up to 40K ctx
1 GPU
RX 7900 XT~87Simulate output on RX 7900 XT
By VRAM tier · typical GPUs and Macs at Q4_K_M
How good is Qwen3 8B?
The llmrun Score merges 7 public benchmark boards into a single figure, where 100 represents the average for open models.
Score vs VRAM in the 24 GB tier
- Qwen3 8B
- Other models
- Best Score for its VRAM
#19 of 36 that fit 24 GB12 of 36 fit your MacBook Air 15" M4 (16 GB) (16 GB)
Each dot is a model at the best quant that fits a 24 GB GPU. The line steps up to the best Score available at each amount of VRAM; a dot below it is beaten by a model that needs no more memory.
Full 24 GB ranking →Benchmarks
The 6 boards that weigh most in the llmrun Score, of 7 with a result for this model; badges show who ran the test.
- GPQA Diamond
Epoch AI · Apr 2025Independent
Independent56.8%
#46/100 open
- AIME 2024/2025
Epoch AI · Apr 2025Independent
Independent56.1%
#42/88 open
- SciCode
Epoch AI · Apr 2025Lab-reported
Lab-reported22.6%
#68/72 open
- CritPt
Epoch AI · Apr 2025Lab-reported
Lab-reported0.0%
#82/84 open
- DTBench
Epoch AI · Apr 2025Lab-reported
Lab-reported59.7%
#52/70 open
- LMCA
Epoch AI · Apr 2025Lab-reported
Lab-reported8.8%
#51/56 open
In the 24 GB class
36 models fit 24 GBSame Score, only models that fit 24 GB at their best quant: the ranking a 24 GB owner actually chooses from.
- #1
Qwen3.8 27BAlibaba
122.0
~38 tok/s · 17.9 GB
- #2
Qwen3.6 27BAlibaba
111.2
~38 tok/s · 17.9 GB
- #3
Qwen3.6 35B A3BAlibaba
106.8
~151 tok/s · 22.0 GB
- #4
Gemma 4 31B ITGoogle
106.7
~32 tok/s · 21.7 GB
- #5
Gemma 4 26B A4B ITGoogle
105.8
~171 tok/s · 16.5 GB
- #6
Qwen3.5 35B A3BAlibaba
105.8
~151 tok/s · 22.0 GB
Alternatives to Qwen3 8B
Qwen 3 family · 79 models →Official models with an llmrun Score, each answering one question; numbers are at the best quant that fits a 24 GB GPU.
Best alternative at this size
Qwen3.5 9B
Alibaba
- 5.5 higher Score
- 1 GB less VRAM
- 16 tok/s slower
Score 99.4 · 6.6 GB · ~103 tok/s
Much faster, similar quality
Qwen3.5 4B
Alibaba
- 4.6 higher Score
- 4.2 GB less VRAM
- 82 tok/s faster
Score 98.5 · 3.4 GB · ~201 tok/s
Other vendor
GPT OSS 20B
OpenAI
- 2.5 higher Score
- 5.5 GB more VRAM
- 71 tok/s faster
Score 96.4 · 13.1 GB · ~190 tok/s
Similar models
10| Model | Score | VRAM | tok/s | Why it is similar |
|---|---|---|---|---|
| Qwen3 14BAlibaba | 98.0 | 11.9 GB | ~69 | similar score |
| Phi 4Microsoft | 83.9 | 12.5 GB | ~69 | similar score |
| Gemma 2 9B ITGoogle | 68.0 | 7.3 GB | ~107 | same size |
| Qwen2.5 7B InstructAlibaba | 65.1 | 5.8 GB | ~131 | same size |
| Llama 3.1 8B InstructMeta | 62.5 | 7.9 GB | ~124 | same size |
| Meta Llama 3 8B InstructMeta | 62.0 | 7.9 GB | ~124 | same size |
| Granite Switch 4.1 8BIBM | — | 8.4 GB | ~111 | same size |
| ALoDLM 8BAmazon | — | 7.6 GB | ~119 | same size |
| Llm Jp 4.1 8B Thinkingllm-jp | — | 7.6 GB | ~115 | same size |
| Ling 3.0 TinyInclusion AI | — | 5.5 GB | ~364 | same size |
Fine-tunes & variants(26)
- ContextPilot 8B
- Goedel Prover v2 8B
- Huihui Qwen3 8B Abliterated v2
- Josiefied Qwen3 8B Abliterated V1
- Molmo2 8B
- Nemotron Orchestrator 8B
- Osim 8B
- Penguin VL 8B
- QiMing Janus 8B
- Qwen Marketing
- Qwen3 8B
- Qwen3 8B Abliterated
- Qwen3 8B Abliterated
- Qwen3 8B F16
- Qwen3 8B Heretic
- Qwen3 8B Heretic
- Qwen3 8B Supermultimodal
- Qwen3 8B UnBias Plus SFT Instruct v2
- Qwen3 8B ZeroZero Heretic
- Qwen3.5 9B Uncensored
- Qwen3Guard Gen 8B
- UIGEN X 8B
- Veritas 8B Fact Checker Non Thinking 1.0
- WebWorld 32B
- WebWorld 8B
- XiaoHong V1
Specifications
- Publisher
- Alibaba
- Family
- Qwen 3
- Parameters
- 8.2B
- Architecture
- Qwen3ForCausalLM
- Context Length
- 40,960 tokens
- Vocabulary Size
- 151,936
- Release Date
- 2025-04-27
- License
- Apache 2.0
Frequently Asked Questions
How much VRAM does Qwen3 8B need?
Qwen3 8B requires 5.5 GB of VRAM at Q4_K_M, or 17.0 GB at BF16. Full 41K context adds up to 5.7 GB (11.3 GB total).
VRAM = Weights + KV Cache + Overhead
Weights = 8.2B × 4.8 bits ÷ 8 = 4.9 GB
KV Cache + Overhead ≈ 0.6 GB (at 2K context + ~0.3 GB framework)
Fit ratings and hardware model lists check this model with room for a 16K-token context, which needs a little more memory.
KV Cache + Overhead ≈ 6.4 GB (at full 41K context)
VRAM usage by quantization
What's the best quantization for Qwen3 8B?
For Qwen3 8B, Q4_K_M (5.5 GB) offers the best balance of quality and VRAM usage. Q4_K_L (5.6 GB) provides better quality if you have the VRAM. The smallest option is IQ2_XXS at 2.9 GB.
VRAM requirement by quantization
★ Recommended — best balance of quality and VRAM usage.
Can I run Qwen3 8B on a Mac?
Yes — Mac Mini M2 Pro (16 GB) and 133 other Macs can run Qwen3 8B. Apple Silicon uses unified memory, so the model shares RAM with the system. At Q4_K_M you need at least 5.5 GB of usable unified memory (RAM minus macOS overhead).
Can I run Qwen3 8B locally?
Yes — Qwen3 8B can run locally on consumer hardware. Popular tools include Ollama, LM Studio, and llama.cpp.
How fast is Qwen3 8B?
At Q4_K_M, Qwen3 8B can reach ~870 tok/s on AMD Instinct MI350X. On NVIDIA GeForce RTX 4090: ~119 tok/s. Speed depends mainly on GPU memory bandwidth. Real-world results typically within ±20%.
tok/s = (bandwidth GB/s ÷ model GB) × efficiency
Example: NVIDIA B200 → 8000 ÷ 5.52 × 0.65 = ~942 tok/s
Estimated speed at Q4_K_M (5.5 GB)
Real-world results typically within ±20%. Speed depends on batch size, quantization kernel, and software stack.
What's the download size of Qwen3 8B?
At Q4_K_M, the download is about 4.91 GB. The full-precision BF16 version is 16.38 GB. The smallest option (IQ2_XXS) is 2.25 GB.
Which GPUs can run Qwen3 8B?
52 consumer GPUs can run Qwen3 8B at Q4_K_M. Top options include AMD Radeon RX 6700 XT, AMD Radeon RX 6800, AMD Radeon RX 6800 XT, AMD Radeon RX 7600. 40 GPUs have plenty of headroom for comfortable inference.
Which devices can run Qwen3 8B?
151 devices with unified memory can run Qwen3 8B at Q4_K_M (5.5 GB), including AMD Ryzen AI 9 HX 370 (Strix Point) Laptop, ASUS Ascent GX10, Beelink GTR9 Pro (Ryzen AI Max+ 395, 128 GB), Framework Desktop (Ryzen AI Max+ 395, 128 GB). Apple Silicon Macs use unified memory shared between CPU and GPU, making them well-suited for local LLM inference.