Configure the scenario Pick a cloud provider, a local rig, and the volume slider. URL updates as you change — share by copying.
§ Cumulative cost over time
cloud burns linearly · local front-loads + amortizes $0 $0 $0 $0 $0 0 3mo 6mo 9mo 12mo 15mo 18mo 21mo 24mo break-even Cloud Local rig Cloud / day: $0.000
Local / day: $0.000
Upfront (est.): $0
Break-even: mo 0
Cloud provider Claude Opus 4.7 ($15/75/M) Claude Sonnet 4.6 ($3/15/M) GPT-5 ($5/20/M) Gemini 2.5 Pro ($1.25/10/M) Claude Haiku 4 ($1/5/M) GPT-5 mini ($0.5/2/M) DeepSeek V3 (API) ($0.27/1.1/M) Llama 3.3 70B (Together) ($0.88/0.88/M) Llama 3.3 70B (Groq) ($0.59/0.79/M) Qwen 3 32B (Together) ($0.6/0.6/M) Qwen 2.5 Coder 32B (DeepInfra) ($0.18/0.18/M)
Horizon 6 months 1 year 2 years 3 years (typical amortization) 5 years
Your local hardware — pick hardware — NVIDIA GB200 NVL72 (13400 GB) AMD Instinct MI355X (288 GB) AMD Instinct MI325X (256 GB) AMD Instinct MI300X (192 GB) NVIDIA B200 (180 GB) NVIDIA H200 (141 GB) NVIDIA H200 NVL (PCIe) (141 GB) AMD Instinct MI250X (128 GB) AMD Instinct MI300A (APU) (128 GB) Intel Gaudi 3 (128 GB) Intel Gaudi 2 (96 GB) NVIDIA RTX PRO 6000 Blackwell (96 GB) NVIDIA H100 NVL (94 GB) NVIDIA A100 80GB SXM (80 GB) NVIDIA H100 PCIe (80 GB) NVIDIA H100 SXM (80 GB) AMD Instinct MI210 (64 GB) NVIDIA RTX 4090 48GB (China-mod) (48 GB) NVIDIA L40 (48 GB) NVIDIA L40S (48 GB) NVIDIA RTX PRO 5000 Blackwell 48GB (48 GB) NVIDIA RTX 6000 Ada Generation (48 GB) NVIDIA RTX A6000 (Ampere) (48 GB) NVIDIA A40 (48 GB) NVIDIA A100 40GB (40 GB) NVIDIA RTX 5000 Ada Generation (32 GB) NVIDIA GeForce RTX 5090 (32 GB) NVIDIA GeForce RTX 5090 Mobile (24 GB) NVIDIA GeForce RTX 4090 (24 GB) NVIDIA GeForce RTX 3090 Ti (24 GB) NVIDIA GeForce RTX 3090 (24 GB) NVIDIA L4 (24 GB) NVIDIA RTX A5000 (24 GB) ASUS ROG Strix Scar 18 (RTX 5090 Mobile) (24 GB) Razer Blade 16 (2025, RTX 5090 Mobile) (24 GB) AMD Radeon RX 7900 XTX (24 GB) NVIDIA RTX 2080 Ti 22GB (China-mod) (22 GB) AMD Radeon RX 7900 XT (20 GB) NVIDIA GeForce RTX 4090 Mobile (16 GB) NVIDIA GeForce RTX 4060 Ti 16GB (16 GB) NVIDIA GeForce RTX 4070 Ti Super (16 GB) NVIDIA GeForce RTX 4080 (16 GB) NVIDIA GeForce RTX 4080 Super (16 GB) NVIDIA GeForce RTX 5060 Ti 16GB (16 GB) NVIDIA GeForce RTX 5070 Ti (16 GB) NVIDIA GeForce RTX 5080 (16 GB) NVIDIA GeForce RTX 3080 16GB (Mobile) (16 GB) AMD Radeon RX 7800 XT (16 GB) Intel Arc A770 16GB (16 GB) AMD Radeon RX 7600 XT (16 GB) AMD Radeon RX 6800 (16 GB) AMD Radeon RX 6800 XT (16 GB) AMD Radeon RX 6900 XT (16 GB) AMD Radeon RX 6950 XT (16 GB) AMD Radeon RX 7900 GRE (16 GB) AMD Radeon RX 9060 XT (16 GB) AMD Radeon RX 9070 (16 GB) AMD Radeon RX 9070 XT (16 GB) NVIDIA GeForce RTX 4070 (12 GB) NVIDIA GeForce RTX 4070 Super (12 GB) NVIDIA GeForce RTX 4070 Ti (12 GB) NVIDIA GeForce RTX 5070 (12 GB) NVIDIA GeForce RTX 3060 12GB (12 GB) NVIDIA GeForce RTX 3080 12GB (12 GB) NVIDIA GeForce RTX 3080 Ti (12 GB) Intel Arc B580 (12 GB) AMD Radeon RX 7700 XT (12 GB) AMD Radeon RX 6700 XT (12 GB) AMD Radeon RX 6750 XT (12 GB) NVIDIA GeForce GTX 1080 Ti (11 GB) NVIDIA GeForce RTX 2080 Ti (11 GB) NVIDIA GeForce RTX 3080 10GB (10 GB) Intel Arc B570 (10 GB) NVIDIA GeForce RTX 4060 (8 GB) NVIDIA GeForce RTX 4060 Ti 8GB (8 GB) NVIDIA GeForce RTX 5060 (8 GB) NVIDIA GeForce GTX 1070 (8 GB) NVIDIA GeForce GTX 1070 Ti (8 GB) NVIDIA GeForce GTX 1080 (8 GB) NVIDIA GeForce RTX 2060 Super (8 GB) NVIDIA GeForce RTX 2070 (8 GB) NVIDIA GeForce RTX 2070 Super (8 GB) NVIDIA GeForce RTX 2080 Super (8 GB) NVIDIA GeForce RTX 3050 (8 GB) NVIDIA GeForce RTX 5060 Ti 8GB (8 GB) NVIDIA GeForce RTX 3060 Ti (8 GB) NVIDIA GeForce RTX 3070 (8 GB) NVIDIA GeForce RTX 3070 Ti (8 GB) Framework Laptop 16 (RX 7700S) (8 GB) AMD Radeon RX 5500 XT 8GB (8 GB) AMD Radeon RX 5700 XT (8 GB) AMD Radeon RX 580 8GB (8 GB) AMD Radeon RX 6600 (8 GB) AMD Radeon RX 6600 XT (8 GB) AMD Radeon RX 6650 XT (8 GB) NVIDIA GeForce GTX 1060 6GB (6 GB) NVIDIA GeForce GTX 1660 (6 GB) NVIDIA GeForce GTX 1660 Super (6 GB) NVIDIA GeForce GTX 1660 Ti (6 GB) NVIDIA GeForce RTX 2060 (6 GB) AMD Radeon RX 5600 XT (6 GB) NVIDIA GeForce GTX 1050 Ti (4 GB) NVIDIA GeForce GTX 1650 (4 GB) NVIDIA GeForce GTX 1650 Super (4 GB) NVIDIA GeForce RTX 3050 Ti (Mobile) (4 GB) AMD Radeon RX 570 (4 GB) NVIDIA GeForce GTX 1060 3GB (3 GB)
Local model (Q4_K_M assumed) — pick model — all-MiniLM-L6-v2 (0.0B) DeepSeek V4 Pro (1.6T MoE) (1600.0B) FLUX.1 [dev] (12.0B) Qwen 3 235B-A22B (235.0B) BGE Large EN v1.5 (0.3B) DeepSeek R1 (671B reasoning) (671.0B) DeepSeek V4 Flash (284B MoE) (284.0B) Kokoro 82M (0.1B) Llama 3.1 8B Instruct (8.0B) Llama 4 Scout (109.0B) Nomic Embed Text v1.5 (0.1B) Qwen 3 0.6B (0.6B) Qwen 3 30B-A3B (30.0B) Llama 3.3 70B Instruct (70.0B) Qwen 2.5 Coder 32B Instruct (32.0B) all-mpnet-base-v2 (0.1B) BGE Reranker v2 M3 (0.6B) Gemma 4 31B Dense (31.0B) Qwen 3 32B (32.0B) XTTS v2 (0.5B) Qwen 3 8B (8.0B) Whisper Base (0.1B) Whisper Small (0.2B) DeepSeek R1 Distill Llama 70B (70.0B) paraphrase-multilingual-MiniLM-L12-v2 (0.1B) Whisper Tiny (0.0B) DeepSeek R1 Distill Qwen 32B (32.0B) DeepSeek V3 (671B MoE) (671.0B) FLUX.1 [schnell] (12.0B) Jina Embeddings v3 (0.6B) Llama 3.2 3B Instruct (3.0B) Multilingual E5 Large Instruct (0.6B) mxbai-embed-large-v1 (0.3B) Qwen 3 1.7B (1.7B) Qwen 3 14B (14.0B) Gemma 3 270M (0.3B) Mistral Small 3 24B (24.0B) Qwen 2.5 7B Instruct (7.0B) Qwen2-VL 2B Instruct (2.0B) DeepSeek R1 Distill Qwen 7B (7.0B) Phi-4 14B (14.0B) Qwen 2.5 14B Instruct (14.0B) Gemma 3 27B (27.0B) Hermes 3 Llama 3.1 8B (8.0B) Llama 3.1 70B Instruct (70.0B) Kimi K2.6 (1000.0B) Mistral Nemo 12B Instruct (12.0B) Phi-4 Reasoning 14B (14.0B) Qwen 2.5 32B Instruct (32.0B) DeepSeek R1 Distill Qwen 14B (14.0B) Llama 3.1 Nemotron 70B Instruct (70.0B) QwQ 32B Preview (32.0B) SigLIP SO400M (patch14-384) (0.9B) Gemma 4 E4B (Effective 4B) (8.0B) Llama 3.2 11B Vision Instruct (11.0B) Distil-Whisper Large v3 (0.8B) Gemma 3 12B (12.0B) Jina Reranker v2 Base Multilingual (0.3B) Qwen 2.5 72B Instruct (72.0B) Qwen 3 4B (4.0B) Qwen3.8 27B (27.0B) SDXL Turbo (2.6B) SmolLM2 135M Instruct (0.1B) Snowflake Arctic Embed L v2.0 (0.6B) TinyLlama 1.1B Chat v1.0 (1.1B) Phi-3.5 Mini Instruct (3.8B) DeepSeek Coder V2 Lite (16B) (16.0B) Gemma 2 2B Instruct (2.6B) Hermes 3 Llama 3.1 70B (70.0B) Llama 3.1 Nemotron Ultra 253B (253.0B) Llama 4 Maverick (400.0B) Pixtral 12B (12.0B) Qwen 3.6 27B (MTP) (27.0B) Trendyol LLM Asure 12B (11.8B) Turkish Gemma 9B T1 (9.0B) Codestral 22B (22.0B) Llama 3.1 Nemotron Nano 8B (8.0B) Llama 3.2 1B Instruct (1.2B) DeepSeek V2 Lite Chat (15.7B) E5 Mistral 7B Instruct (7.1B) Florence-2 Large (0.8B) Gemma 2 9B Instruct (9.0B) Gemma 3 4B (4.0B) GTE ModernBERT Base (0.1B) Mistral 7B Instruct v0.3 (7.0B) Mistral Large 2 (123B) (123.0B) Trendyol LLM 7B Chat v0.1 (7.0B) Turkish Llama 8B Instruct v0.1 (8.0B) Gemma 4 E2B (Effective 2B) (5.1B) Dolphin 3.0 Mistral 24B (24.0B) Omni 31B Turkish Reasoning (31.0B) Cosmos Llama 3 8B Turkish (8.0B) Gemma 4 Turkish 26B (4B active) (20.4B) Kumru 2B (2.4B) Llama 3.2 90B Vision Instruct (90.0B) Mixtral 8x7B Instruct (47.0B) Stable Diffusion 3.5 Medium (2.5B) Command R+ 104B (104.0B) mxbai-rerank-large-v2 (1.5B) Phi-3.5 Vision (4.2B) Piper (0.0B) Ring-2.6-1T (1000.0B) Turkcell LLM 7B v1 (7.4B) Command R 35B (35.0B) Mixtral 8x22B Instruct (141.0B) Parakeet TDT 0.6B v2 (0.6B) Qwen 3.5 2B Turkish SFT (2.0B) YTU Turkish Gemma 9B v0.1 (9.2B) Gemma 3 1B (1.0B) Kanarya 2B (2.0B) Yi 1.5 34B (34.0B) MedGemma 27B (27.0B) SmolLM2 360M Instruct (0.4B) CodeGemma 7B (8.5B) F5-TTS (0.3B) GOT-OCR 2.0 (0.6B) Kanarya 750M (0.8B) Trendyol LLM 7B Base v0.1 (7.0B) VBART Large (Turkish Summarization) (0.4B) ColPali v1.3 (3.0B) SmolVLM Instruct (2.3B) Turkish GPT-2 Large (0.8B) Command R7B (12-2024) (8.0B) Orpheus 3B 0.1 FT (3.8B) OLMo 2 1B Instruct (1.5B) Mistral Turkish v2 (brooqs) (7.2B) Granite 3.1 2B Instruct (2.5B) Malhajar Mistral 7B Turkish (7.2B) RefinedNeuro RN TR R2 (8.0B) RefinedNeuro RN TR R1 (8.0B) Falcon 3 3B Instruct (3.2B) OpenELM 3B Instruct (3.0B) Aya 23 35B (35.0B) Aya 23 8B (8.0B) Aya Expanse 32B (32.0B) BGE M3 (0.6B) CodeQwen 1.5 7B (7.0B) Codestral Mamba 7B (7.0B) Command R+ (Aug 2024) (104.0B) DeepSeek Coder V2 236B (236.0B) DeepSeek MoE 16B Base (16.0B) DeepSeek R1 Distill Llama 8B (8.0B) DeepSeek R1 Distill Qwen 1.5B (1.8B) DeepSeek V2.5 236B (236.0B) Dolphin 3.0 Llama 3.2 3B (3.0B) EXAONE 3.5 2.4B (2.4B) EXAONE 3.5 32B (32.0B) Falcon 3 10B (10.0B) Falcon 3 7B Instruct (7.0B) Falcon Mamba 7B (7.0B) GLM-4 9B (9.0B) GLM-4V 9B (13.9B) Granite 3.0 2B Instruct (2.5B) Granite 3.0 8B Instruct (8.0B) Granite 3.2 8B (8.0B) Granite 3.3 8B (8.0B) Hermes 3 Llama 3.2 3B (3.0B) Hunyuan Large 389B MoE (389.0B) InternLM 2.5 7B Chat (7.0B) InternLM 3 8B (8.0B) InternVL 2.5 26B (26.0B) InternVL 2.5 78B (78.0B) Jamba 1.5 Large (398.0B) Jamba 1.5 Mini (52.0B) Janus-Pro 7B (7.0B) Llama 3.2 11B Vision (11.0B) Llama 3.2 90B Vision (90.0B) LLaVA 1.6 Mistral 7B (7.0B) LLaVA-OneVision 7B (7.0B) MiniCPM 3 4B (4.0B) MiniCPM-V 2.6 8B (8.0B) Ministral 8B Instruct (8.0B) Molmo 72B (72.0B) Molmo 7B-D (8.0B) Moondream 2 (1.9B) Nemotron Mini 4B Instruct (4.0B) NV-Embed v2 (7.8B) OLMo 2 13B (13.7B) OpenBioLLM Llama 3 70B (70.0B) OpenCoder 8B (8.0B) PaliGemma 2 10B (10.0B) PaliGemma 2 3B (3.0B) Phind CodeLlama 34B v2 (34.0B) Qwen 2-VL 7B (7.0B) Qwen 2.5 0.5B Instruct (0.5B) Qwen 2.5 1.5B Instruct (1.5B) Qwen 2.5 3B Instruct (3.0B) Qwen 2.5 Coder 1.5B (1.5B) Qwen 2.5 Coder 14B Instruct (14.0B) Qwen 2.5 Coder 3B (3.0B) Qwen 2.5 Math 72B (72.0B) Qwen 2.5 Math 7B (7.0B) Qwen 2.5-VL 3B (3.0B) Qwen 2.5-VL 72B (72.0B) Qwen 2.5-VL 7B (7.0B) Qwen 3 Embedding 8B (8.0B) RWKV 7 'Goose' 1.5B (1.5B) SmolLM 2 1.7B Instruct (1.7B) SmolLM 2 360M Instruct (0.4B) SmolLM 3 3B (3.0B) Stable LM 2 12B (12.0B) StarCoder 2 15B (16.0B) StarCoder 2 3B (3.0B) StarCoder 2 7B (7.0B) Tulu 3 70B (70.0B) Tulu 3 8B (8.0B) Whisper Large v3 (1.6B) Whisper Large v3 Turbo (0.8B) Yi Coder 9B (9.0B)
Price figures are our planning estimates (launch MSRP, or a used-market estimate for older cards), not live prices. Check the current price before you buy.
Local (per day)
$0
pick rig
Upfront (est.)
$0
hardware only
Break-even
immediate
local beats cloud
§ Verdict at 500,000 tokens/day
Pick a local hardware + model to see the crossover analysis.
Rent instead
Rent the GPU, not the API
A third option between an API and buying hardware: run open-weight models yourself on a GPU you rent by the hour, and pay only while the instance is up. Privacy depends on the host you pick, so read its terms before you send anything sensitive.
Referral links: RunLocalAI may earn a commission or account credits if you sign up and rent, at no extra cost to you. Hourly rates and which GPUs are listed change all the time, so check the provider's page before you start. How we make money .