Runs comfortablyAlibaba
Qwen3.5 9B
Chat · Reasoning · Coding · Vision · Apache-2.0
Parameters 9B
Estimated memory 6.9 GBQ4_K_M
Estimated speed 24–36 tok/sMLX
Arena open-source rank —No exact Arena match
Free local LLM compatibility checker
Can I Run AI Locally on My PC, GPU, or Mac?
Run LLMs locally with fewer guesses. Check popular local AI models against your hardware, then compare VRAM, memory, quantization, context length, and estimated tokens per second before you download.
Free to use. No software install. Hardware detection runs in your browser, and raw GPU details are not uploaded.
Your hardware
Detecting hardware…
Unified memory48 GBSystem RAM48 GBBandwidth273 GB/s
Quantization / precision
Context
31Comfortable
0Tight
3Offload
5Won't fit
39 models evaluated
What Local LLM Can I Run on This Hardware?
Search local AI models for chat, coding, reasoning, and vision. Compare compatibility, estimated memory, speed, runtime, and the evidence behind each result.
Task
Compatibility
Provider
Sort
Arena ranking is shown for 20 models that can be matched exactly; unmatched models stay unranked rather than being guessed.
Arena · CC BY 4.0 · 2026-07-10Runs comfortablyAlibabaMoE
Qwen3 30B-A3B
Chat · Reasoning · Coding · Apache-2.0
Parameters 31B3.3B active
Estimated memory 20 GBQ4_K_M
Estimated speed 65–98 tok/sMLX
Arena open-source rank #821317 · 26,474 votes
Runs comfortablyAlibabaMoE
Qwen3.5 35B-A3B
Chat · Reasoning · Coding · Vision · Apache-2.0
Parameters 35B3B active
Estimated memory 23 GBQ4_K_M
Estimated speed 72–108 tok/sMLX
Arena open-source rank #451396 · 29,184 votes
Runs comfortablyMeta
Llama 3.1 8B Instruct
Chat · Coding · Llama 3.1 Community
Parameters 8B
Estimated memory 6.3 GBQ4_K_M
Estimated speed 27–40 tok/sMLX
Arena open-source rank #1301187 · 49,605 votes
Runs comfortablyOpenAIMoE
GPT-OSS 20B
Chat · Coding · Reasoning · Apache-2.0
Parameters 21B3.6B active
Estimated memory 14 GBQ4_K_M
Estimated speed 60–90 tok/sMLX
Arena open-source rank #931288 · 10,621 votes
Runs comfortablyAlibaba
Qwen2.5 Coder 7B Instruct
Coding · Chat · Apache-2.0
Parameters 7.6B
Estimated memory 6 GBQ4_K_M
Estimated speed 28–43 tok/sMLX
Arena open-source rank —No exact Arena match
Runs comfortablyDeepSeek
DeepSeek R1 Distill Qwen 14B
Reasoning · Chat · MIT
Parameters 15B
Estimated memory 11 GBQ4_K_M
Estimated speed 15–22 tok/sMLX
Arena open-source rank —No exact Arena match
Runs comfortablyAlibaba
Qwen3 8B
Chat · Reasoning · Coding · Apache-2.0
Parameters 8.2B
Estimated memory 6.4 GBQ4_K_M
Estimated speed 26–39 tok/sMLX
Arena open-source rank —No exact Arena match
Runs comfortablyAlibaba
Qwen2.5 7B Instruct
Chat · Reasoning · Apache-2.0
Parameters 7.6B
Estimated memory 6 GBQ4_K_M
Estimated speed 28–43 tok/sMLX
Arena open-source rank —No exact Arena match
Runs comfortablyDeepSeek
DeepSeek R1 Distill Qwen 7B
Reasoning · Chat · MIT
Parameters 7.6B
Estimated memory 6 GBQ4_K_M
Estimated speed 28–43 tok/sMLX
Arena open-source rank —No exact Arena match
Runs comfortablyDeepSeek
DeepSeek R1 Distill Qwen 32B
Reasoning · Chat · MIT
Parameters 33B
Estimated memory 23 GBQ4_K_M
Estimated speed 6.6–9.9 tok/sMLX
Arena open-source rank —No exact Arena match
Runs comfortablyGoogle
Gemma 3 12B IT
Chat · Reasoning · Vision · Gemma Terms
Parameters 12B
Estimated memory 8.8 GBQ4_K_M
Estimated speed 18–27 tok/sMLX
Arena open-source rank #751334 · 3,829 votes
Runs comfortablyAlibaba
Qwen2.5 Coder 14B Instruct
Coding · Chat · Apache-2.0
Parameters 15B
Estimated memory 11 GBQ4_K_M
Estimated speed 15–22 tok/sMLX
Arena open-source rank —No exact Arena match
Runs comfortablyAlibaba
Qwen3 14B
Chat · Reasoning · Coding · Apache-2.0
Parameters 15B
Estimated memory 11 GBQ4_K_M
Estimated speed 15–22 tok/sMLX
Arena open-source rank —No exact Arena match
Runs comfortablyGoogle
Gemma 3 27B IT
Chat · Reasoning · Vision · Gemma Terms
Parameters 27B
Estimated memory 19 GBQ4_K_M
Estimated speed 8–12 tok/sMLX
Arena open-source rank #621358 · 47,514 votes
Runs comfortablyMicrosoft
Phi-4 14B
Chat · Coding · Reasoning · MIT
Parameters 14B
Estimated memory 10 GBQ4_K_M
Estimated speed 15–23 tok/sMLX
Arena open-source rank #1211217 · 24,126 votes
Runs comfortablyAlibaba
Qwen2.5 Coder 32B Instruct
Coding · Chat · Apache-2.0
Parameters 33B
Estimated memory 22 GBQ4_K_M
Estimated speed 6.6–10 tok/sMLX
Arena open-source rank #1131230 · 5,432 votes
Runs comfortablyMistral AI
Mistral Small 3.1 24B
Chat · Coding · Vision · Apache-2.0
Parameters 24B
Estimated memory 17 GBQ4_K_M
Estimated speed 9–14 tok/sMLX
Arena open-source rank #991278 · 33,199 votes
Runs comfortablyAlibaba
Qwen2.5 14B Instruct
Chat · Reasoning · Apache-2.0
Parameters 15B
Estimated memory 11 GBQ4_K_M
Estimated speed 15–22 tok/sMLX
Arena open-source rank —No exact Arena match
Runs comfortablyAlibaba
Qwen3 32B
Chat · Reasoning · Coding · Apache-2.0
Parameters 33B
Estimated memory 23 GBQ4_K_M
Estimated speed 6.6–9.9 tok/sMLX
Arena open-source rank #711340 · 3,926 votes
Runs comfortablyGoogle
Gemma 3 4B IT
Chat · Vision · Gemma Terms
Parameters 4B
Estimated memory 3.7 GBQ4_K_M
Estimated speed 54–81 tok/sMLX
Arena open-source rank #911291 · 4,171 votes
Runs comfortablyAlibaba
Qwen2.5 32B Instruct
Chat · Reasoning · Apache-2.0
Parameters 33B
Estimated memory 22 GBQ4_K_M
Estimated speed 6.6–10 tok/sMLX
Arena open-source rank —No exact Arena match
Runs comfortablyMicrosoft
Phi-4 Mini Instruct
Chat · Coding · Reasoning · MIT
Parameters 3.8B
Estimated memory 3.6 GBQ4_K_M
Estimated speed 57–85 tok/sMLX
Arena open-source rank —No exact Arena match
Runs comfortablyMeta
Llama 3.2 3B Instruct
Chat · Llama 3.2 Community
Parameters 3B
Estimated memory 3 GBQ4_K_M
Estimated speed 72–108 tok/sMLX
Arena open-source rank #1561110 · 7,936 votes
Runs comfortablyMistral AI
Mistral 7B Instruct v0.3
Chat · Coding · Apache-2.0
Parameters 7.3B
Estimated memory 5.8 GBQ4_K_M
Estimated speed 30–44 tok/sMLX
Arena open-source rank —No exact Arena match
Runs comfortablyGoogle
Gemma 2 9B IT
Chat · Gemma Terms
Parameters 9.2B
Estimated memory 7 GBQ4_K_M
Estimated speed 24–35 tok/sMLX
Arena open-source rank #1221208 · 54,611 votes
Runs comfortablyMicrosoft
Phi-3.5 Mini Instruct
Chat · Coding · MIT
Parameters 3.8B
Estimated memory 3.6 GBQ4_K_M
Estimated speed 57–85 tok/sMLX
Arena open-source rank —No exact Arena match
Runs comfortablyMistral AI
Mistral NeMo 12B Instruct
Chat · Coding · Apache-2.0
Parameters 12B
Estimated memory 8.9 GBQ4_K_M
Estimated speed 18–27 tok/sMLX
Arena open-source rank —No exact Arena match
Runs comfortablyGoogle
Gemma 2 27B IT
Chat · Reasoning · Gemma Terms
Parameters 27B
Estimated memory 19 GBQ4_K_M
Estimated speed 8–12 tok/sMLX
Arena open-source rank #1121232 · 75,754 votes
Runs comfortablyMeta
Llama 3.2 1B Instruct
Chat · Llama 3.2 Community
Parameters 1B
Estimated memory 1.7 GBQ4_K_M
Estimated speed 200–300 tok/sMLX
Arena open-source rank #1831055 · 8,045 votes
Runs comfortablyZhipu AI
GLM-4 9B Chat
Chat · Coding · GLM-4 License
Parameters 9.4B
Estimated memory 7.2 GBQ4_K_M
Estimated speed 23–34 tok/sMLX
Arena open-source rank —No exact Arena match
Runs with offloadMeta
Llama 3.1 70B Instruct
Chat · Reasoning · Llama 3.1 Community
Parameters 70B
Estimated memory 47 GBQ4_K_M
Estimated speed 0.5–0.7 tok/sMLX
Arena open-source rank #1071261 · 55,240 votes
Runs with offloadMeta
Llama 3.3 70B Instruct
Chat · Reasoning · Llama 3.3 Community
Parameters 70B
Estimated memory 47 GBQ4_K_M
Estimated speed 0.5–0.7 tok/sMLX
Arena open-source rank #1001275 · 54,726 votes
Runs with offloadDeepSeek
DeepSeek R1 Distill Llama 70B
Reasoning · Chat · MIT + Llama
Parameters 70B
Estimated memory 47 GBQ4_K_M
Estimated speed 0.5–0.7 tok/sMLX
Arena open-source rank —No exact Arena match
Won't fitOpenAIMoE
GPT-OSS 120B
Chat · Coding · Reasoning · Apache-2.0
Parameters 117B5.1B active
Estimated memory 75 GBQ4_K_M
Estimated speed 1.7–2.5 tok/sMLX
Arena open-source rank #591366 · 30,628 votes
Won't fitMoonshot AIMoE
Kimi K2 Instruct
Chat · Coding · Reasoning · Modified MIT
Parameters 1,000B32B active
Estimated memory 633 GBQ4_K_M
Estimated speed 0.3–0.4 tok/sMLX
Arena open-source rank —No exact Arena match
Won't fitMetaMoE
Llama 4 Scout 17B-16E
Chat · Reasoning · Vision · Llama 4 Community
Parameters 109B17B active
Estimated memory 70 GBQ4_K_M
Estimated speed 0.5–0.8 tok/sMLX
Arena open-source rank #981281 · 30,280 votes
Won't fitMiniMaxMoE
MiniMax M2.1
Chat · Coding · Reasoning · MiniMax Open License
Parameters 230B10B active
Estimated memory 146 GBQ4_K_M
Estimated speed 0.9–1.3 tok/sMLX
Arena open-source rank —No exact Arena match
Won't fitMetaMoE
Llama 4 Maverick 17B-128E
Chat · Reasoning · Vision · Llama 4 Community
Parameters 400B17B active
Estimated memory 254 GBQ4_K_M
Estimated speed 0.5–0.8 tok/sMLX
Arena open-source rank #921288 · 39,963 votes
Catalog 2026.07.17-mvp.2Updated 2026-07-17Engine 1.1.0
Performance is an estimate based on public specifications. Drivers, runtime versions, cooling, context length, and background load can change real results.
More than fit or fail
Local LLM Compatibility With VRAM, Memory, and Speed Explained
Use the free compatibility checker as an LLM VRAM calculator and decision guide—not just a hardware list. Every estimate keeps its inputs and limits visible.
PC, GPU, and Mac Compatibility
Start with browser detection, choose a known GPU or Mac profile, or edit VRAM, unified memory, system RAM, and bandwidth yourself.
LLM VRAM and Memory Calculator
Estimate model weights, context KV cache, and runtime overhead instead of comparing a model file size with advertised VRAM alone.
Quantization That Matches Your Hardware
Compare Q4, Q5, Q8, BF16, FP8, and NVFP4 while the checker explains memory tradeoffs and hardware-specific precision support.
Estimated Local LLM Speed
See an estimated output speed range in tokens per second. It is a planning estimate, not a claim about a benchmark on your exact machine.
Reasons and Sources, Not a Black Box
Open a result to inspect its memory breakdown, fit reason, runtime, source, catalog date, and evaluation engine version.
Browser-Local Hardware Detection
Automatic detection runs in your browser. Raw GPU adapter details are not uploaded, and you can always replace an approximate result manually.
Before you download
Run LLM Locally: A Practical Hardware and Memory Checklist
Running LLM locally takes more than matching parameter count to advertised VRAM. Check usable memory, model format, quantization, context, and runtime overhead before choosing a download.
Check my local LLM setupStart With Usable VRAM or Unified Memory
Leave room for the operating system, display, inference runtime, model context, and temporary buffers instead of treating every advertised gigabyte as available.
Choose a Local LLM Model and Quantization
Match the model architecture and parameter count with a practical Q4, Q5, Q8, BF16, FP8, or NVFP4 build that your hardware and runtime support.
Use the LLM Calculator Before Downloading
Compare estimated weights, KV cache, runtime overhead, fit status, and tokens per second before committing storage space and setup time.
How to run an LLM locally
Check Local LLM Hardware Requirements in Three Steps
Start with your PC, GPU, or Mac; adjust model precision and context; then use the fit, memory, and speed estimates to decide what to download.
Check my hardware now1. Confirm your PC, GPU, or Mac
The checker tries WebGPU and WebGL locally. Browser privacy limits can make detection approximate, so you can select a profile or edit the hardware values.
2. Set quantization and context
Q4, Q5, and Q8 work broadly; BF16 needs more memory; FP8 and NVFP4 require specific NVIDIA architectures. Longer context increases KV-cache memory.
3. Choose a local AI model
Use fit status, memory breakdown, estimated speed, runtime, source, catalog version, and engine version together—not a single VRAM number.
Local LLM FAQ
Can I Run AI Locally? Frequently Asked Questions
Practical answers about local LLM hardware requirements, VRAM, Mac unified memory, quantization, speed estimates, and browser-based detection.
Find the Local LLM Your Hardware Can Actually Run
Use the free local LLM compatibility checker before downloading model files. Compare fit, VRAM, memory, quantization, context, and estimated speed in one place.