# Keep a rarely used tuned model callable without paying for idle time

> Keep a rarely used tuned model callable without paying for idle time: 0 companies fully do it, 6 with a limit, 10 carry the tag. Every listing in Fine-tuning with the reason and a source.

- Canonical: https://www.anchorterminal.com/niches/fine-tuning.idle-serving-cost/
- Markdown: https://www.anchorterminal.com/niches/fine-tuning.idle-serving-cost/index.md (~1,100 tokens)
- Slim: https://www.anchorterminal.com/niches/fine-tuning.idle-serving-cost/index.min.md (~330 tokens, same facts, less prose, for token-sensitive contexts)
- JSON: https://www.anchorterminal.com/niches/fine-tuning.idle-serving-cost/index.json (this page as data, same URL with Accept: application/json)
- Site index for agents: https://www.anchorterminal.com/llms.txt (the key docs and every listing in one file, at most 16 MB, about 4 million tokens: https://www.anchorterminal.com/llms-full.txt)
- API: https://www.anchorterminal.com/api/v1/index.json
- MCP: https://www.anchorterminal.com/mcp (the whole catalogue as tools, no key)
- DNS: dig +short TXT <slug>.grade.anchorterminal.com (the grade as a DNS TXT record, for a sandbox with no HTTP where names still resolve; also top., best.<category>. and <a>.vs.<b>. under the zone, and the same under grade.letme.dev)
- Licence: CC BY 4.0, credit Anchor Terminal and link this page (https://www.anchorterminal.com/terms/#data)
- Updated: 2026-11-15

No company's docs show a full fit yet. The listings below do the job with a limit, or not at all.

| Companies, full fit | With a limit | Carry the tag | Share of the category's work | First-hand accounts |
| --- | --- | --- | --- | --- |
| 0 | 6 | 10 | 12 | 6 |

## What counts as done

The docs state the hourly hosting fee, storage fee and idle-deletion rule for a tuned model, and either offer per-token serving with no idle charge or a deploy that can be scripted and is ready within a stated time.

How agents ask for it: what does it cost to keep my fine-tune deployed; serve my LoRA per token; can I undeploy it when it's not in use

## Does the job, with a limit

It does the job, but something in the docs or a user's account will get in an agent's way. The limit is named in each row.

- **Amazon Bedrock model customisation (Amazon Web Services, Inc.)**: On-demand deployment bills only for use and is created by API with a client request token, but it covers four Nova models and Llama 3.3 70B; acc-25 found imported models keep billing storage after clean-up.
- **Microsoft Foundry fine-tuning (Azure OpenAI) (Microsoft Corporation)**: The cost guide states the hourly hosting fee (720 hours a month in its example) and that Developer deployments vanish after 24 hours, but no idle-free per-token option lasts, and acc-26 and acc-27 found no pause and minutes-long redeploys.
- **Fireworks AI Fine-tuning (Fireworks.ai, Inc.)**: Serverless LoRA has no deployment to pay for and loads in under five minutes, but only for GLM 5.3 Flash; every other tuned model goes to on-demand GPUs billed by the hour.
- **Nebius Token Factory fine-tuning (Nebius B.V.)**: Tuned models run only on dedicated endpoints, which bill while replicas are ready; the hosting price sits in a console page robots.txt disallows, so the docs state no hourly fee or idle rule.
- **Together AI Fine-tuning (Together Computer, Inc.)**: The docs state that dedicated inference bills per minute per running replica even when idle and that scaling to zero stops billing, but no per-token serving exists for tuned models and no time to scale back up was found.
- **Vertex AI Gemini tuning (Google LLC)**: Tuned endpoints bill per prediction token with no hosting fee on the pricing page, but neither page read states a storage fee or whether idle tuned endpoints are deleted.

## Couldn't check

The docs were closed to us, or silent. These carry no mark either way.

- **Tinker (Thinking Machines Labs, Inc.)**: Sampling and checkpoint storage prices are on the docs site, which was not read because the Service Terms bar automated scraping of the documentation.

- not this job: Axolotl (Axolotl AI): Not a hosting service.
- not this job: Unsloth (Unsloth): Not a hosting service; serving cost depends on wherever the export runs.

## What stops agents finishing

- hourly hosting fee charged whether or not the model is called
- deployment deleted after an idle period
- redeploy takes minutes, too slow for on-demand use

## Around it

Before: fine-tuning.managed-job. After: inference.tool-loop.

## Where this comes from

Weight 12 of 100 in Fine-tuning: 6 of 47 accounts: two Microsoft Q&A threads on Azure hosting fees, a Bedrock imported-model storage issue, a Tinker user for whom the economics failed, and two Together users citing serverless LoRA at the base price

As data: https://www.anchorterminal.com/api/v1/use-cases/fine-tuning.idle-serving-cost.json, or the MCP tool get_use_case with id fine-tuning.idle-serving-cost. Draft, read 2026-11-15 from public docs and first-hand accounts, not yet reviewed by a person.

---

Query this catalogue directly over MCP: https://www.anchorterminal.com/mcp (Streamable HTTP, no key). How: https://www.anchorterminal.com/agents/#mcp
