What is Tensorfuse?
Serverless GPUs on your own cloud for fine-tuning and inference.
Who is Tensorfuse best for?
Tensorfuse is best for AI teams that want serverless GPU inference and fine-tuning inside their own AWS account.
Teams that must keep models in their own cloud
Models run as autoscaling endpoints inside your AWS account, and datasets and weights stay in your own S3 bucket.
Fine-tuning open models
Train open-source models on your own data with Axolotl, Unsloth or Hugging Face, or bring your own training loop.
Apps with many LoRA adapters
The Growth plan serves thousands of LoRA adapters from a single GPU, hot-swapping them as requests arrive.
ML engineers working from a laptop
Dev containers connect local code to cloud GPUs and sync every change, with no SSH or manual setup.
How much does Tensorfuse cost?
Tensorfuse's Hacker plan is free with 100 managed GPU hours, and its paid plans cost $249 a month for Starter or $799 a month for Growth, on top of what you pay your own cloud for the GPUs.
- 100 managed GPU hours (MGH)
- Serverless inference and dev containers
- Community support
- 2,000 MGH, then $0.10 per MGH
- Adds fine-tuning, GitHub Actions and custom domains
- Private Slack support, 14-day free trial
- 5,000 MGH, then $0.10 per MGH
- Adds batch jobs, job queues, environments and multi-LoRA inference
- Premium support, 14-day free trial
Billed annually, 15% less: Starter is $212 and Growth $679 a month. An MGH (managed GPU hour) is Tensorfuse's fee on top of your cloud bill: an A100 your provider charges $4 an hour for costs $4.10 an hour in total. Enterprise, priced on request, adds custom MGH with volume discounts, role-based access control, SSO, SOC 2 and HIPAA security and dedicated engineering support. Seed-stage startups with under $500K in funding may qualify for 10,000 free managed GPU hours over 6 months. Full pricing on tensorfuse.io. We checked these prices on .
How do readers rate Tensorfuse?
No one has rated Tensorfuse yet. Be the first.
Sign in to rate. One rating per account.
What are the best Tensorfuse alternatives?
One tool on Recaply does Tensorfuse's job. It is open source, so you can run it on your own server.
Questions about Tensorfuse
What is Tensorfuse?
Serverless GPUs on your own cloud for fine-tuning and inference.
Who is Tensorfuse for?
Tensorfuse is best for AI teams that want serverless GPU inference and fine-tuning inside their own AWS account.
Is Tensorfuse free?
Yes. Tensorfuse has a free Hacker plan: 100 managed GPU hours (MGH).
How much does Tensorfuse cost?
Tensorfuse's paid plans are Starter at $249 per month and Growth at $799 per month.
Is there an open source alternative to Tensorfuse?
Yes. dstack is an open source tool that does the same job, so you can run your own.
What are the best Tensorfuse alternatives?
The closest alternatives on Recaply are dstack.
Where can I try Tensorfuse?
On its official website, tensorfuse.io.
How does Tensorfuse charge for GPU hours?
Tensorfuse runs on your own cloud account, so you pay your provider for the GPUs and Tensorfuse charges for the hours it manages. Plans include 100, 2,000 or 5,000 managed GPU hours, and Starter and Growth add $0.10 for each hour after that.
Together AI
Modal
Cloudflare
Salad
Modular
Upstash