Niches · Fine-tuning

Run a tuning job through a hosted API and call the tuned model

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

0companies, full fit
5companies, fit with a limit
13companies carry the tag
19share of the category's work
10first-hand accounts

What counts as done

The job is created, polled and cancelled by API with a documented status and event stream, the dataset is validated with errors that name the bad row, any safety screening of data or of the tuned model is documented before submission, and the finished model id is callable on the same inference API as base models.

Measured, for our own tests: for 1,000 examples on a 7-8B model, the job reaches a callable model within 4 hours of creation. No docs we read state these figures, so they don't count against a listing here.

How agents ask for it: fine-tune a model on this dataset, start a tuning job with these examples, train a LoRA on Qwen with our data and give me the model id

Does the job, with a limit 5

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.

ListingWhy
Amazon Bedrock model customisation A
Amazon Web Services, Inc.
CreateModelCustomizationJob takes a clientRequestToken and returns a jobArn to monitor or stop, and on-demand deployments are called by ARN on the runtime API; but data and roles live in S3 and IAM first (acc-24), and no data or model safety screening is described in the pages read. source
Microsoft Foundry fine-tuning (Azure OpenAI) C
Microsoft Corporation
OpenAI-shaped create, cancel, events and checkpoint calls on /openai/v1, and data safety checks are not billed; but the safety criteria are not described in the pages read, and the tuned model must be deployed with a Foundry Owner role before it can be called. source
Nebius Token Factory fine-tuning D
Nebius B.V.
OpenAI-style create, get, cancel, events and checkpoint endpoints with terminal statuses, but no data or model safety screening is documented, and serving the result needs custom model weights on a dedicated endpoint, a beta available on request. source
Together AI Fine-tuning C
Together Computer, Inc.
Jobs move pending, queued, running, uploading, completed with an events endpoint and a local check_file before upload, but there is no idempotency key, no data or model safety screening is documented, and the tuned model is served only after a dedicated endpoint is deployed. source
Vertex AI Gemini tuning B
Google LLC
tuningJobs create, get and cancel with JOB_STATE values, and the tuned model gets an endpoint on the same prediction API, but no dataset error format or safety screening was found in the pages read, and acc-20 to acc-22 hit documented calls returning errors, 404 and a rejected tuned-model name. source

Doesn't do this job 4

Listings in the category that do something else, with the reason.

ListingWhy
Axolotl B
Axolotl AI
A command-line trainer on the owner's GPUs; there is no hosted job API, status page or callable model id. source
Fireworks AI Fine-tuning B
Fireworks.ai, Inc.
Managed SFT and DPO no longer accept new jobs (jobs created before 9 November 2026 keep their models), and managed RFT closed on 1 October; the Training API replacing them runs the training loop in the caller's own Python process. source
Tinker D
Thinking Machines Labs, Inc.
The user writes and runs the training loop against the API; there is no submit-and-poll job with dataset validation, per the cookbook README and acc-36. source
Unsloth C
Unsloth
A local library, Studio and desktop app; there is no hosted job API. source

What stops agents finishing

  • job sits in validation or queue for hours with no status detail
  • trained model fails a post-training safety eval after training was billed
  • IAM role or region setup a desk agent cannot do alone

Around it

Before: agent-observability.offline-eval-regression, file-storage.upload. After: inference.tool-loop, agent-observability.offline-eval-regression.

Where this comes from

Weight 19 of 100 in Fine-tuning: 10 of 47 accounts, the largest group: Vertex SDK issues, a Bedrock workshop issue, a Together integration plan, and OpenAI forum threads on jobs stuck in validation or failed by safety checks; share of 52 account-to-use-case links, rounded

The same as data: /api/v1/use-cases/fine-tuning.managed-job.json, or the MCP tool get_use_case with id fine-tuning.managed-job. Other jobs in this category: /api/v1/niches?category=fine-tuning.

Draft, read 2026-11-15 from public docs and first-hand accounts, not yet reviewed by a person. A mark says what the docs show, not what a run measured.

For companies

Do agents find, use and choose your tools?

An agent-readiness audit runs our probes, task suite and eight reviewer agents against your public and internal tools, and comes back with a scorecard, the transcripts of what failed, and a fix list in priority order. From $2,500, re-run included. We never take payment to move a rank. We do help companies earn one.