Free
Free
Free software plan; GPU/storage billed separately
- Allowance and restrictions
- One import and training run/month; one model; 1,000 requests/month; no vision.
Tensorant pricing
● Identity checkedOpen-model fine-tuning and inference · Official-source review 2026-10-09 UTC
Tensorant connects dataset preparation, LoRA/QLoRA training, base-versus-tuned evaluation and model serving. Training and serving GPUs run in the user’s RunPod account, with data on their storage volume.
- Category
- Open-model fine-tuning and inference
- Free access
- Free orchestration plan for one monthly import, one monthly training run and one published model; RunPod GPU and storage charges are separate.
- API access
- OpenAI-compatible chat-completions endpoint, streaming and scoped API keys; standard LoRA adapter download and Hugging Face publication documented.
Understand the total cost
Tensorant pricing & plans
Starter
$99.00 /mo
See stated billing terms
- Allowance and restrictions
- Unlimited imports; five training runs/month; five models; 50,000 requests; five members.
Pro
$499.00 /mo
See stated billing terms
- Allowance and restrictions
- Vision enabled; five million requests/month; ten seats; $30/month per extra seat; underlying GPU/storage extra.
Price history
No retained pricing changes yet. A current price alone does not establish a historical trend.
How this profile is supported
Official product information
Product identity and relevant official-source evidence reviewed individually.
Read original source ↗Costs and constraints
Free includes 1,000 requests/month, 30/minute and four concurrent requests. Starter is $99/month for five training runs, five published models, 50,000 requests and five members. Pro is $499/month for vision workflows, five million requests and ten members; extra seats cost $30/month. GPU time, persistent storage and outside endpoints are billed separately. Monthly counters reset on the first UTC day; failed requests reaching a model count. Standard plans cap published models and API keys at 1,000 even where labelled unlimited. Cancellation/downgrade retains existing material but prevents additions above the lower limit. No training or inference benchmark was performed.
Read original source ↗Facts apply to the named version and check date. Send a sourced correction if something changed.