localLLMrig
Your prompt picks the LLM. The LLM picks the hardware. Test the AI first. Buy the hardware after.
Compare local LLMs against Claude, GPT, Gemini and Grok on your own prompts.
1 · Prompt Write a Python function tha… 2 · Local LLM Qwen2.5 32B 3 · Hardware GeForce RTX 3090 24GB
1 · Prompt
Example “Write a Python function that removes duplicate rows from a CSV
11 words · ~21 tokens
Read as Coding
2 · Local LLM
Local models for coding
Qwen2.5 32B Top pick Qwen2.5 14B Llama 3.3 70B
Our pick Qwen2.5 32B is the mid-size Qwen instruct model in this catalog. Source
Q4 at 8,192 tokens: 22.6 GB VRAM. Calculator estimate.
Prompt → LLM → Hardware Prompt → LLM LLM → Hardware
Try an example: Trick question: bat and ball Plan 3 days in Lisbon on €500 Translate EN → NL · FR · DE
Choose LLMs or ✦ Suggest for my question Suggest 4 ticked, up to 8 at the same time Split by the hardware you need to run each model yourself. Groups in dollars, machine prices in euros. Free: tick 2 to 8 per question.
Tick the free models to test Hardware under $1,5006 free · 18 with credits Small PC, desktop or Mac. From the Mac mini M4 16GB ($615 €549 Mac mini M6 32GB ($1,416 €1,530
Gemma 4 31BFree Often busy Gemma · 31B Qwen3.8 27B Qwen · 27.78B Gemma 4 26B A4BFree Often busy Gemma · 26B, 4B active gpt-oss-20bCannot read images OpenAI · 21B, 3.6B active Nemotron 3.5 LightningFree Cannot read images Nemotron · 30B, 3B active Llama 3.1 8B InstructCannot read images Meta · 8.03B Qwen3 Coder 30B A3BCannot read images Qwen · 30B, 3B active Nemotron 3 Nano OmniFree Nemotron · 30B, 3B active Qwen3.5 9B Qwen · 9B Mistral Small 3.2 24B Mistral · 24B North Mini CodeFree Cannot read images North · 30B, 3B active LFM2.5 24B A2BCannot read images Liquid AI · 24B, 2.45B active GLM 4.7 FlashCannot read images Z.ai · 30B, 3B active Xing4.0 29BFree Cannot read images XingChen · 29B, 4B active Qwen3 14BCannot read images Qwen · 14B Phi-4Cannot read images Microsoft · 14.7B Gemma 3 27B Google · 27B Gemma 3 12B Google · 12B Ministral 3 14B Mistral · 14B Nemotron 3 Nano 30B A3BCannot read images NVIDIA · 30B, 3.5B active Llama 3.2 3B InstructCannot read images Meta · 3.21B Gemma 3 4B Google · 4B Ministral 3 3B Mistral · 3B Llama 3.2 1B InstructCannot read images Meta · 1B Hardware over $5,0002 free · 18 with credits Pro workstation or company server. From the Mac Studio M5 Ultra 256GB ($11,565 €12,500 source ) for the largest.
DeepSeek V4.1 Flash 1017 DeepSeek · 763.21B, 16B active Laguna S 2.1Free Cannot read images Laguna · 118B, 8B active MiniMax M3.1 Flash MiniMax · 428B, 23B active Dots3-Note PreviewFree Dots · 280B, 16B active DeepSeek R1 0528Cannot read images DeepSeek · 671B, 37B active Qwen3.8 Flash Next Qwen · 180B, 6B active GLM-5.3 Flash Z.ai · 320B, 18B active Meta Muse Spark Open Meta · 816B, 42B active GLM-5.3Cannot read images Z.ai · 753B, 40B active DeepSeek V4 Pro 0813Cannot read images DeepSeek · 1600B, 49B active Kimi K2.6 Moonshot AI · 1000B, 32B active Nemotron 3 Ultra 550B A55BCannot read images NVIDIA · 550B, 55B active Qwen3 235B A22BCannot read images Qwen · 235B, 22B active MiniMax M3 MiniMax · 428B, 23B active MiMo V2.6 Flash Xiaomi · 309B, 15B active DeepSeek V3Cannot read images DeepSeek · 671B, 37B active Step 5 Preview StepFun · 600B, 27B active Step 3.7 Flash StepFun · 198B, 11B active Hunyuan Hy4 PreviewCannot read images Tencent · 770B, 49B active Hunyuan Hy3Cannot read images Tencent · 295B, 21B active Online only18 with credits GPT, Claude, Gemini, Grok and the largest open models. At most 3 of these per question: they cost the most to run, and one question on credits may cost at most a few cents.
Claude Sonnet 5.5 Anthropic GPT-6.1 Sol OpenAI Gemini 4 Argon Google Grok 4.7 xAI Claude Opus 5.5 Anthropic Qwen3.8 Max Qwen GPT-6 Luna OpenAI Gemini 3.8 Flash Google Kimi K3.1 Pro Moonshot AI · 2800B · too big for one machine Kimi K3 Moonshot AI · 2800B, 104B active · too big for one machine GLM-5.3 PrimeCannot read images Z.ai Mistral Large 4 Mistral AI GPT-5.6 Terra OpenAI Gemini 3.1 Pro Preview Google Claude Haiku 5.5 Anthropic GPT-5.6 Sol OpenAI Meta Muse Spark 1.3 Meta Command A Plus Cohere Free needs no credits; a model without a tag uses credits. Cannot read images