Simple pricing for your AI models
Fine-tune models on your own hardware for free. When you're ready to put a model online, FineTuneMyAI can run it on managed GPU infrastructure.
Local Fine-Tuning
Your hardware computesFine-tune open-weight LLMs on your own Mac or PC without paying software licensing or cloud markups.
What is included for $0:
- LoRA & QLoRA 4-bit fine-tuning: Train 7B to 70B models with minimal memory.
- Dataset preparation & validation: JSONL syntax checks and quality heuristics.
- Apple Silicon & NVIDIA acceleration: Native MLX Metal and CUDA 12 support.
- Local model evaluation: Perplexity scoring and side-by-side prompt comparisons.
- On-device data sovereignty: Datasets, weights, and adapters never leave your drive.
- Export & local inference: Run your trained adapters locally in Ollama or vLLM.
No credit card required. No subscription required.
Hosted Model Inference
Managed GPU infrastructureDeploy your fine-tuned model to dedicated cloud GPUs. Pay only for the compute time you use.
Inference capabilities:
- 1-Click deployment: Turn any trained LoRA adapter into an OpenAI-compatible REST API.
- Per-minute metering: Pay strictly for running compute. No upfront commitments.
- Automatic scale-to-zero: Endpoints scale down when idle to eliminate wasted spend.
- Transparent price estimates: Review estimated hourly costs before launching.
- Dedicated VRAM allocations: Fast token generation with low TTFT latency.
- Optional private VPC: Secure enterprise deployment with TLS and IP allowlists.
No upfront fees. Pay only when your model is serving inference.
Hosted Inference Compute Tiers
Initial indicative compute rates for hosting fine-tuned models on FineTuneMyAI-managed infrastructure.
Up to ~8B parameters
Llama 3.1 8B, Qwen 2.5 7B, Mistral 7B
Optimized for small, highly efficient domain adapters and lightweight microservices.
~9B to 14B parameters
Qwen 2.5 14B, Gemma 2 9B, Mistral Nemo 12B
Balanced capacity for specialized code reasoning, multilingual tasks, and extraction.
~15B to 32B parameters
Qwen 2.5 32B, Command R 35B
High-precision reasoning, mathematical proofs, and complex agentic tool use.
~33B to 70B parameters
Llama 3.3 70B, Qwen 2.5 72B
Frontier-grade open model performance for enterprise-grade autonomous reasoning.
70B+ or high-concurrency clusters
DeepSeek 67B+, multi-instance distributed endpoints
Dedicated multi-GPU nodes with guaranteed SLAs, private VPC peering, and custom quantization.
Model size is a useful starting point, but actual compute requirements depend on quantization, context length, neural architecture, and expected concurrent traffic. FineTuneMyAI recommends the appropriate compute tier and calculates the estimated hourly cost before you confirm any deployment.
Fine-tune locally. Deploy when you need it.
Fine-Tune Locally
Train on your GPU for $0
Evaluate Accuracy
Verify loss & metrics locally
Choose Deployment
Run locally or send to cloud
Review Estimated Cost
Approve hourly GPU rate
Launch Endpoint
Auto-scales and idles to zero
Frequently Asked Pricing Questions
Is FineTuneMyAI free to use for fine-tuning?
Yes. FineTuneMyAI is currently free for local fine-tuning. You use your own compatible hardware (Apple Silicon Mac or NVIDIA GPU) to perform all training computations. Datasets, weights, and adapter checkpoints remain strictly on your machine with zero subscription fees.
Do I need a paid subscription to use FineTuneMyAI?
No. There is no software subscription or license fee required to fine-tune models locally. You can prepare datasets, run LoRA and QLoRA tuning, evaluate models, and manage your local model library for $0.
What do users pay for?
Users pay only when they choose to deploy and host their fine-tuned model on FineTuneMyAI-managed GPU infrastructure for production inference. You are billed strictly for the GPU compute time consumed by your deployed endpoint.
Can I run my fine-tuned model locally instead of paying for hosting?
Absolutely. Every fine-tuned adapter and merged model produced by FineTuneMyAI is yours to keep. You can run it locally for free on Ollama, MLX, vLLM, or LM Studio without ever deploying to the cloud.
How does billing work for hosted inference?
Hosted inference uses transparent, pay-as-you-go compute pricing. Before launching any endpoint, FineTuneMyAI inspects your model requirements and displays the estimated hourly cost. Endpoints support automatic scale-down and scale-to-zero when idle so you are never charged for inactive resources.
Does my dataset leave my computer when fine-tuning?
Never during local fine-tuning. Training corpora, extracted text, raw weights, and LoRA adapters remain on your workstation drive. The cloud control plane receives only high-level progress telemetry (loss floats, step counts, and token counts). When you choose hosted inference, only the final model artifact needed to serve requests is uploaded to the inference cluster.
Is FineTuneMyAI open source?
No. FineTuneMyAI is proprietary software built on a local-first architecture. It coordinates with open-weight base models and transparent local runtimes, but the platform software itself is proprietary.