Qwen3.5 4B Instruct (Q8)
Local OKExcellentA · 40
Best for: Coding, Agents, Multimodal·Pop: 88/100
Perf: est. 47 tok/s, typical error 5% · first token ~1.6s
Fits in 8 GB VRAM with room to spare. Best for coding, agents, multimodal on RTX 4060.
Best Local AI Models for RTX 4060 (8GB)
The RTX 4060 is the most affordable current-gen NVIDIA GPU, but its 8GB VRAM is the real constraint for local AI. It runs 7B-8B models at Q4 with little room for context, landing around est. 40 tok/s, typical error 5%. For anything bigger, the 16GB RTX 4060 Ti or a rented cloud GPU is the smarter path.
Quick answer
The best local LLM for the RTX 4060 is Qwen3.5 4B Instruct (Q8) (est. 47 tok/s, typical error 5%) on its 8GB VRAM. It uses ~4.3GB of VRAM; the RTX 4060 handles up to 8B parameter models at Q4. A 14B model at Q4 with CPU offload runs at est. 5 tok/s, typical error 15%. (ModelFit, 2026)
Sizing rule: a Q4 model needs about 0.6 GB of VRAM per billion parameters, and ModelFit budgets 90% of this card's 8GB for weights, context, and KV-cache. The per-size table below uses that same budget. Other strong fits: Ornith 1.0 9B (9B, ~5.6GB) and LFM2.5 8B-A1B (8.3B, ~5.5GB). On this card Qwen3.5 4B Instruct (Q8) runs at est. 47 tok/s, typical error 5%. What it will not run: a 14B model does not fit in VRAM alone, so it spills to system RAM over PCIe and slows sharply.
$ollama run qwen3.5:4b-q8_0
- Top pick
- Qwen3.5 4B Instruct (Q8)
- Decode speed
- est. 47 tok/s, typical error 5%
- VRAM needed
- ~4.3 GB
Speeds are ModelFit estimates from memory bandwidth and model size, not measured benchmarks. Typical error: the median gap between the ModelFit speed model and measured runs on hardware the model was not fitted on, for this backend and model size.
Cite this page: ModelFit, RTX 4060 8GB Local LLM (2026): What 8GB VRAM Runs, est. 40 tok/s, https://modelfit.io/gpu/rtx-4060/, updated October 2026, CC BY 4.0.
Specs
VRAM
8 GB GDDR6
Speed (8B Q4)
est. 40 tok/stypical error 5%
Bandwidth
272 GB/s
Architecture
Ada Lovelace
Max model size
Up to 8B parameter models
Compatibility
10 excellent, 0 workable
Vendor
Table 1
RTX 4060 estimated tokens/sec by model size
Q4_K_M · ModelFit estimate| Model size | Est. speed | Fit on 8GB |
|---|---|---|
| 7B | est. 46 tok/stypical error 5% | Fits in VRAM |
| 14B | est. 5 tok/stypical error 15% | CPU offload (slow) |
| 20B MoE (3.6B active) | est. 19 tok/stypical error 15% | CPU offload (slow) |
| 32B | est. 1 tok/stypical error 5% | CPU offload (slow) |
| 35B MoE (3B active) | est. 23 tok/stypical error 15% | CPU offload (slow) |
| 70B | est. 0 tok/stypical error 10% | CPU offload (slow) |
| 120B MoE (5.1B active) | est. 14 tok/stypical error 15% | CPU offload (slow) |
Fig. 1RTX 4060: estimated speed by model size
Decode speed, generic Q4_K_M model of each size at 4K context, 272 GB/s, typical error 5-15%
Fits in 8 GBCPU offload (slow)
7B46
14B5
20B MoE (3.6B active)19
32B1
35B MoE (3B active)23
70B0
120B MoE (5.1B active)14
tok/s est.ModelFit estimates, not measured benchmarks: decode speed of a generic Q4_K_M model of each size at 4K context on the RTX 4060's 272 GB/s bandwidth. MoE rows scale by active parameters (decode reads only the active experts), so a 35B MoE runs far faster than a dense 32B. "CPU offload" sizes exceed the 8GB VRAM; dense models slow to a crawl there, MoE models degrade less because hot experts stay GPU-resident. What typical error means.
Fig. 2
Context costs VRAM too.
Qwen3.5 4B Instruct (Q8) loads ~4.3 GB of weights; at 16k context the KV cache adds ~0.5 GB (still fits the ~7 GB usable VRAM), and at 64k it adds ~2.0 GB (still fits).
Weights 4.3 GBKV cache · 16k context 0.5 GBHeadroom 2.4 GBRuntime reserve 0.8 GB
The model leaves about 2.4 GB of the usable VRAM budget free after weights and 16k context.
KV-cache figures assume an fp16 cache, the llama.cpp/Ollama default. Standard GQA models use a size-class estimate (8 KV heads x 128 head dim class); hybrid linear-attention models (Qwen3.5/3.6, Qwen3-Next) use the exact per-token cost from their published config, since only their sparse full-attention layers cache KV. A q8_0 KV cache roughly halves either figure. Estimates, not measurements.
Get told when a model beats Qwen3.5 4B Instruct (Q8) on your RTX 4060
One email when a new open-weight model is a better fit for 8 GB of VRAM, plus the Thursday weekly.
You'll also get the weekly ModelFit email. By signing up you agree to our Privacy Policy. Unsubscribe from either email anytime, in one click.
Where to buy the RTX 4060
Storage & accessories for your model library
- Internal NVMe SSD · 2TBA Gen4 M.2 drive keeps your whole GGUF and quant collection on fast local storage, loading models straight off NVMe.Check price on Amazon
- USB4 NVMe Enclosure40Gbps external storage fast enough to run models from. Pair it with an M.2 drive for a portable model vault.Check price on Amazon
ModelFit may earn a commission on purchases through these links, at no extra cost to you.
RTX 4060 VRAM for AI: what actually fits?
8GB GDDR6 at 272 GB/s holds one 7B-8B model at Q4 quantization: Qwen 2.5 7B or Llama 3.2 8B use ~5-5.6GB, leaving only 2-3GB for the KV cache and context. That caps practical context length and rules out 14B models without heavy CPU offloading (which cuts speed 70-80%). The RTX 4060 is a fine entry point for small models, but 8GB is the spec that frustrates you first. If you expect to run 14B models, start with the RTX 4060 Ti 16GB instead.
Table 2
What does not fit in 8GB (and what it costs you)
With 8GB, the honest ceiling is a single 7B model at Q4. Anything larger does not simply run slower: the layers that do not fit spill to system RAM over PCIe, and decode speed falls off a cliff. These are the same bandwidth-derived estimates used in the table above, shown here as the penalty rather than the headline.
| Model size | Fits in 8GB? | Est. speed | Slowdown vs 7B |
|---|---|---|---|
| 7B | Yes | est. 46 tok/stypical error 5% | n/a |
| 14B | No, needs ~2GB more in system RAM | est. 5 tok/stypical error 15% | ~9x slower |
| 20B MoE (3.6B active) | No, needs ~5GB more in system RAM | est. 19 tok/stypical error 15% | ~2x slower |
| 32B | No, needs ~12GB more in system RAM | est. 1 tok/stypical error 5% | ~46x slower |
| 35B MoE (3B active) | No, needs ~14GB more in system RAM | est. 23 tok/stypical error 15% | ~2x slower |
Mixture-of-experts models are the exception worth knowing: they activate only a fraction of their parameters per token, so a large MoE can stay usable on a small card where a dense model of the same total size will not. Speeds are ModelFit estimates from memory bandwidth and model size, not measured benchmarks.
Table 3
RTX 4060 vs similar GPUs
| Hardware | Memory | Speed | Bandwidth |
|---|---|---|---|
| RTX 4060 | 8 GB | est. 40 tok/stypical error 5% | 272 GB/s |
| RTX 3070 | 8 GB | est. 61 tok/stypical error 5% | 448 GB/s |
| RTX 3060 | 12 GB | est. 50 tok/stypical error 5% | 360 GB/s |
| RTX 4060 Ti | 16 GB | est. 43 tok/stypical error 5% | 288 GB/s |
| RTX 5060 Ti | 16 GB | est. 68 tok/stypical error 5% | 448 GB/s |
Recommended models
registry-verified10 modelsOrnith 1.0 9B
Local OKOKB · 22
Best for: Agentic coding on small machines·Pop: 76/100
Perf: est. 43 tok/s, typical error 5% · first token ~0.6s
Fits in 8 GB VRAM with room to spare. Best for agentic coding on small machines on RTX 4060.
LFM2.5 8B-A1B
Local OKOKB · 24
Best for: On-device agents, tool calling, multilingual chat·Pop: 72/100
Perf: est. 118 tok/s, typical error 15% · first token ~0.1s
Fits in 8 GB VRAM with room to spare. Best for on-device agents, tool calling, multilingual chat on RTX 4060.
Gemma 4 E4B
Local OKExcellentA · 44
Best for: On-device, Mobile, Chat·Pop: 82/100
Perf: est. 61 tok/s, typical error 5% · first token ~0.5s
Fits in 8 GB VRAM with room to spare. Best for on-device, mobile, chat on RTX 4060.
MiMo-V2.6-Distill-Qwen-9B
Local OKOKB · 19
Best for: Agentic coding, Small machines·Pop: 60/100
Perf: est. 42 tok/s, typical error 5% · first token ~0.6s
Fits in 8 GB VRAM with room to spare. Best for agentic coding, small machines on RTX 4060.
Gemma 4 E2B (Q8)
Local OKExcellentA · 36
Best for: IoT, Mobile, Edge·Pop: 76/100
Perf: est. 97 tok/s, typical error 5% · first token ~1.6s
Fits in 8 GB VRAM with room to spare. Best for iot, mobile, edge on RTX 4060.
Qwen3.5 9B Instruct
Local OKOKB · 3
Best for: Quality, Coding, Reasoning·Pop: 86/100
Perf: est. 43 tok/s, typical error 5% · first token ~0.6s
Fits in 8 GB VRAM with room to spare. Best for quality, coding, reasoning on RTX 4060.
Granite 4.2 8B
Local OKExcellentA · 31
Best for: Chat, Coding·Pop: 66/100
Perf: est. 38 tok/s, typical error 5% · first token ~0.6s
Fits in 8 GB VRAM with room to spare. Best for chat, coding on RTX 4060.
Ministral 3 8B
Local OKOKB · 17
Best for: Chat, Coding, Vision·Pop: 74/100
Perf: est. 36 tok/s, typical error 5% · first token ~0.6s
Fits in 8 GB VRAM with room to spare. Best for chat, coding, vision on RTX 4060.
Qwen3.5 4B Instruct
Local OKExcellentA · 51
Best for: Coding, Agents, Multimodal·Pop: 88/100
Perf: est. 75 tok/s, typical error 5% · first token ~0.3s
Fits in 8 GB VRAM with room to spare. Best for coding, agents, multimodal on RTX 4060.
Video and image models on the RTX 4060
Runs · Stated: the linked source says a GPU with that much VRAM runs the model. Runs · Published: a dated run on the named card with that much VRAM. unknown: no source yet. Exceeds VRAM: the complete download is larger than 8 GB, so part of it has to sit in system RAM. Every VRAM tier: video · image.
Video
| Model | Fits | Quality | App |
|---|---|---|---|
| LTX-2.3LTX-2 Community License Agreement | Runs · PublishedRTX 4060 · 8 GB VRAM · ComfyUI GGUF workflow, 720p 5 s in about 190 s, 32 GB RAM · ComfyUI issue #14345 | No independent score yet | Lightricks/LTX-2 ltx-pipelines, ComfyUI |
| MiniMax H3MiniMax H3 Community License Agreement | Runs · PublishedRTX 4050 · 6 GB VRAM · ComfyUI int8 convrot, 0.4 megapixel, 5 s, 16 GB RAM · ComfyUI issue #15482 | #8 of 48 in text to video11,417 votessnapshot 2026-09-22 | stable-diffusion.cpp, ComfyUI |
| Wan2.2 A14BApache License 2.0 | Runs · PublishedRTX 2060 · 6 GB VRAM · ComfyUI Q8 GGUF, 480x720, 81 frames, 7 to 10 min, 28 GB RAM plus a 60 GB page file · ComfyUI issue #12451 | #42 of 48 in text to video10,458 votessnapshot 2026-09-22 | ComfyUI |
| Wan2.2 TI2V 5BApache License 2.0 | Runs · Stated8 GB VRAM · ComfyUI native offloading · ComfyUI docs: Wan2.2Complete: Comfy-Org repackaged FP16 with FP8 scaled UMT518.1 GB download | No independent score yet | ComfyUI |
| Mochi 1 previewApache License 2.0 | unknownComplete: calcuis GGUF Q3_K_M with Q4_0 T5 and FP8 VAE7.7 GB download | #48 of 48 in text to video5,893 votessnapshot 2026-09-22 | Diffusers, ComfyUI |
Quality: blind human votes from Arena (arena.ai), CC BY 4.0, snapshot 2026-09-22. Scores are for the Arena checkpoint, not your local quantized build.
Image
| Model | Fits | Quality | App |
|---|---|---|---|
| FLUX 3 Image KleinFLUX Kommunity License v1.0 | Runs · Stated8 GB VRAM · official ComfyUI int8 files · FLUX 3 Image Klein README (Black Forest Labs) | No independent score yet | Diffusers |
| FLUX.1 [dev]FLUX.1 [dev] Non-Commercial License v1.1.1 | Runs · Stated6 GB VRAM · stable-diffusion.cpp with GGUF weights · stable-diffusion.cpp FLUX docs | #78 of 82 in text to image49,221 votessnapshot 2026-10-07 | Diffusers, ComfyUI |
| Qwen-Image-2.1Qwen Research License Agreement | Runs · Stated7 GB VRAM · DiffSynth-Studio, weights streamed from disk · DiffSynth-Studio docs: Qwen-Image-2.1 | #19 of 82 in text to image11,441 votessnapshot 2026-10-07 | Diffusers, stable-diffusion.cpp |
| SDXL Base 1.0CreativeML Open RAIL++-M License | Runs · Stated8 GB VRAM · SDXL 1.0, no app named · SDXL 1.0 announcement (Stability AI) | No independent score yet | ComfyUI, stable-diffusion.cpp |
| Z-Image-TurboApache 2.0 | Runs · Stated4 GB VRAM · stable-diffusion.cpp with GGUF weights · stable-diffusion.cpp Z-Image docs | #59 of 82 in text to image24,808 votessnapshot 2026-10-07 | Diffusers, stable-diffusion.cpp |
| FLUX.2 [dev]FLUX Non-Commercial License | Exceeds VRAMComplete: Q4_K_M GGUF34.8 GB download | #38 of 82 in text to image77,988 votessnapshot 2026-10-07 | Diffusers, ComfyUI, stable-diffusion.cpp |
| FLUX.2 [klein]Apache 2.0 (4B), FLUX Non-Commercial License (9B) | Exceeds VRAMComplete: Klein 4B Q8_0 GGUF12.5 GB download | No independent score yet | Diffusers, stable-diffusion.cpp |
| Krea 2 TurboKrea 2 Community License Agreement v.1 | Exceeds VRAMComplete: Turbo Q4_K_M GGUF16.9 GB download | No independent score yet | Diffusers, stable-diffusion.cpp |
| Stable Diffusion 3.5 LargeStability AI Community License | Exceeds VRAMComplete: Q4_0 GGUF16.4 GB download | #81 of 82 in text to image23,396 votessnapshot 2026-10-07 | ComfyUI |
Quality: blind human votes from Arena (arena.ai), CC BY 4.0, snapshot 2026-10-07. Scores are for the Arena checkpoint, not your local quantized build.
Can I run it on the RTX 4060?
Per-model verdict pages for this exact card: weights, KV cache by context length, and estimated tok/s, each derived from the same engine the recommendations above use.
- Qwen3 8B Q4_K_Mfits
- Qwen3 14B Q4_K_Mno
- GPT-OSS 20B MXFP4no
- Qwen3.8 27B Q4_K_Mno
- Llama 3.3 70B Instruct Q4_K_Mno
- Qwen3.5 35B-A3B Instruct Q4_K_Mno
- Gemma 4 26B-A4B Q4_K_Mno
- Qwen3.5 4B Instruct Q4_K_Mfits
- Gemma 4 E4B Q4_K_Mfits
- Gemma 3 4B Instruct Q4_K_Mfits
- Llama 3.1 8B Instruct Q4_K_Mfits
- Qwen3.5 9B Instruct Q4_K_Mfits
- Gemma 4 12B Q4_K_Mtight
- Mistral Nemo 12B Q4_K_Mtight
- Qwen3.5 9B Instruct (Q8) Q8_0tight
- Gemma 4 12B (Q8) Q8_0no
- LFM2 24B-A2B Instruct Q4_K_Mno
- Qwen3.5 27B Instruct Q4_K_Mno
- Qwen3.6 27B Q4_K_Mno
- Gemma 4 31B Q4_K_Mno
- Qwen3.6 35B-A3B Q4_K_Mno
- Qwen3.8 27B (Q8) Q8_0no
Similar GPUs for local AI
- RTX 4060 Ti16GB · est. 43 tok/s
- RTX 306012GB · est. 50 tok/s
- RTX 30708GB · est. 61 tok/s
- RTX 5060 Ti16GB · est. 68 tok/s
Speeds are estimates for Qwen3 8B Q4_K_M, typical error 5%.
Compatible model families
- QwenAlibaba Cloud: Qwen3.8 27B local flagship (Apache 2.0, vision)
- LlamaMeta: Most popular open-weight model family
- DeepSeekDeepSeek AI: V4.1 Flash: multimodal 552B MoE, API only
- MistralMistral AI: Excellent performance-per-parameter ratio
- GemmaGoogle DeepMind: Gemma 4 E4B: 4.5B effective, runs on 8GB Macs
- PhiMicrosoft: Best quality-per-gigabyte at small sizes
RTX 4060 FAQ: common questions
How much VRAM does the RTX 4060 have for LLMs?
The RTX 4060 has 8GB GDDR6 VRAM. After driver and OS overhead, about 7.5GB is usable for model loading. That fits a single 7B-8B model at Q4 quantization, but leaves limited room for long context windows.
What size LLM can I run on an RTX 4060?
Up to 8B parameters at Q4 quantization, and even then context is tight. Good picks are Qwen 2.5 7B (~5.2GB), Llama 3.2 8B (~5.6GB), and Mistral 7B (~4.4GB). 14B models require CPU offloading, which makes them very slow.
Is the RTX 4060 good for local AI in 2026?
It works for 7B-8B models, but 8GB VRAM is limiting. The 2026 shortage has compressed the gap to the RTX 4060 Ti 16GB, which or a used RTX 3060 12GB gives meaningfully more headroom. Buy the 4060 only if you already own it or are on a strict budget.
RTX 4060 vs RTX 4060 Ti for running LLMs?
The RTX 4060 Ti 16GB is the better AI card: double the VRAM lets it run 14B models the base 4060 cannot. The 4060 is fine for 7B models, but if local AI is your goal, the 16GB 4060 Ti is worth the extra cost.
How fast is a 27B-class model on the RTX 4060?
The RTX 4060's 8GB of VRAM cannot fit a 32B model comfortably. The largest size class it fits is 7B, at est. 46 tok/s, typical error 5%. The largest individual model in the catalog that fits is Ornith 1.0 9B (9B).
Related guides & benchmarks
- How to Install Ollama: Complete Setup GuideStep-by-step Ollama installation for beginners on any platform.
- Qwen 3.5 Small Models: 4B Beats 20BSmall models that run great on 8GB GPUs like the RTX 4060.
- Local LLMs vs GPT-4 and Claude: How They CompareSee how local 7B-8B models on your GPU compare to cloud APIs.
More local AI guides
- How Much RAM (or VRAM) Do You Need for a Local LLM?The model-size-to-memory matrix: what each VRAM and RAM tier actually runs.
- Best LLM for MacBook (Apple Silicon)Unified-memory Macs run bigger models per dollar. Picks by RAM tier, M1 to M5.
- Local AI image generator on MacCompare image models by RAM and app.
- Local image generation on MacImage models and apps by Mac RAM.
Browse all NVIDIA GPUs for AI
Real-world calibration
Have this hardware? Share a measured run.
Time a model that fits your machine with Ollama, using the same prompts as every other run. The CLI shows your result and asks before it sends anything; your machine then joins the speed test leaderboard. It does not change the quality benchmark.
$ npx @modelfit/cli@latest bench
Want personalized recommendations?
Use our interactive wizard to compare models across Apple Silicon and NVIDIA GPUs.