OPEN MODELS / Language models (LLM)
Run reflection:70b on a cloud GPU
reflection:70b is a Language models (LLM) model by Ollama. See how much GPU memory it needs, which GPUs it fits on today and what it costs to run it yourself, billed per second.
From the project: A high-performing model trained with a new technique called Reflection-tuning that teaches a LLM to detect mistakes in its reasoning and correct course.
GPUs it fits on today
About 47 GB of GPU memory: the model download (4-bit by default) plus working memory.
| GPU | Memory | Per hour | 10 hours | |
|---|---|---|---|---|
| RTX A6000 | 48 GB | $0.72 | $7.20 | Rent |
| CMP 170HX | 64 GB | $0.75 | $7.50 | Rent |
| L40 | 48 GB | $0.87 | $8.70 | Rent |
| RTX 6000Ada | 48 GB | $1.00 | $10.00 | Rent |
| L40S | 48 GB | $1.03 | $10.30 | Rent |
Live starting prices from our catalog. You pay per second while the machine runs.
How to run reflection:70b
- 01
Rent a GPU
Pick one of the GPUs above. Billing is per second; stop the machine and you stop paying.
- 02
Open a terminal or Jupyter
Connect over SSH or open Jupyter in your browser. PyTorch and CUDA are ready to use.
- 03
Install Ollama and run the model
Run the commands below. Ollama downloads the model and serves an OpenAI-compatible API on port 11434.
curl -fsSL https://ollama.com/install.sh | sh
ollama run reflection:70bWhy run it yourself?
Your own LLM server has no per-token bill and no rate limit, and your data stays private. Pick a GPU by memory and serve the model with vLLM.
Questions
- How much GPU memory does reflection:70b need?
- By our estimate about 47 GB for inference (70B parameters). Higher resolution, longer context or bigger batches need more.
- How much does it cost to run reflection:70b?
- From $0.72 per hour on RTX A6000, billed per second. Ten hours cost about $7.20.
- Do I need a subscription?
- No. You pay only for the time your machine runs, from your USD balance. Stop it whenever you are done.
Try it on your own GPU in minutes
Pick a GPU, open Jupyter or SSH and run any open model. You pay per second and can stop anytime; no subscription, no setup fee.
- Per-second billing in USD
- PyTorch and CUDA ready in minutes
- Rent from ChatGPT or Claude too