Microsoft Foundry fine-tuning (Azure OpenAI) by Microsoft Azure

# Fix list: Microsoft Foundry fine-tuning (Azure OpenAI)

From Anchor Terminal's listing at https://www.anchorterminal.com/tools/azure-foundry-fine-tuning, the October 2026 research run, assessed 1 October 2026. Grade C, 61.4 out of 100.

This is everything the published grade says the listing lacks, the biggest possible gain to the total first. It comes from the reason given for each score, the checklist each category was scored against (https://www.anchorterminal.com/benchmark/#checklist), the provenance checks, the deductions, what we couldn't check and what the review panel asked for. A fix counts at the next check, once it's public.

For a coding agent working on Microsoft Foundry fine-tuning (Azure OpenAI): work through the items below in the product, its docs and its public pages. Each category gives the reason for its score, with the points each checklist item earned, and the checklist itself, so the gap is the items that earned less than their points. Change the product, not the wording, and keep a note of what you changed and where it's published.

## 1. Payments & pricing, 20 out of 100, up to 10 more on the total

Why it scored 20: No machine payment protocol (0). Per-1M-token training rates and the $1.70 hourly hosting fee are public in the Azure Retail Prices API without a login, even though the pricing page's table needs a browser (20). Azure's free account needs a card (0). An Azure subscription and resource must exist before any call (0).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-payments):

The published rubric, also on the [x402 page](https://www.anchorterminal.com/x402/).

- 40, a machine payment protocol (x402, MPP or L402) on the tool's own endpoints. 10 to 30 when it covers only some endpoints or only goes through a third party, and the note says which.
- 20, per-call or per-unit pricing published without a login. 10 for public plan-only pricing, 0 for "contact sales" or prices behind a login.
- 20, a free tier or trial that doesn't need a card.
- 20, autonomous onboarding, meaning an agent can get access without a person signing up in a browser (keyless use, x402, a programmatic key API).

Payment platforms and agent wallets rarely charge for their own API over a machine protocol, so the first line has steps for them, and the highest one that applies counts. 40 when x402, MPP or L402 runs on all their own endpoints, 30 when it runs on part of their own API, 25 when their merchants can accept one, 20 for running a facilitator, 15 for paying as a buyer, and 0 when the only protocol is their own. Merchant acceptance sits above a facilitator because the platform's own customers can charge agents through it, while a facilitator settles for sellers who wire up the protocol themselves. The counter-argument (a facilitator does more for the protocol as a whole) has a point. Each note says which step applied.

Open-source software you run yourself is scored on its hosted or paid option if it has one. A free, self-hosted package with nothing to buy gets 20, 20 and 20 for the last three lines, and 0 to 40 for the first only if it ships a payment protocol.

## 2. Agent ergonomics, 47 out of 100, up to 8.6 more on the total

Why it scored 47: Job objects are compact and list calls take `limit`; no field selection (10). Cursor pagination with `after` on the OpenAI-shaped lists; no filters found (10). Errors come back in the OpenAI error shape with a code and message; no fine-tuning error table (12). No idempotency key on job creation (5). Only `model` and `training_file` are required and the openai SDK exists in several languages, but deploying the result needs a second call to the Resource Manager API under a different role (10).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-ergonomics):

- 0 to 25, context cost. For MCP, the number and size of the tool definitions (25 for ten or fewer compact tools, 15 for 11 to 30, 5 for more than 30, plus up to 10 back for toolsets, dynamic loading or read-only subsets). For APIs, whether responses can be sized (field selection, limits, summaries).
- 20, pagination, filtering and output-size controls.
- 20, actionable, documented error responses, codes and messages an agent can recover from.
- 20, idempotency or safe retries, and for MCP the `readOnlyHint` and `destructiveHint` annotations.
- 15, sensible defaults, few required parameters, and official SDKs in at least two languages.

Models are read for tool use, structured output, prompt caching, context length, batch and SDKs. Frameworks for how much code and how many defaults a tool-calling agent with MCP needs.

## 3. Reliability, 65 out of 100, up to 7 more on the total

Why it scored 65: Azure status page with post-incident reviews kept for five years and Azure OpenAI Service as a listed component (20). One major in the window, 29 September 2026, when Azure OpenAI, Foundry Models and Cognitive Services saw intermittent failures and higher latency in Sweden Central, one of the three regional training regions, for about six hours (10). Fine-tuning limits published with numbers, 3 simultaneous training jobs (5 on the developer tier), 20 queued, 100 jobs and 100 files per resource, 2 billion tokens per job and 720 hours (15). The quota page says to retry with backoff and links code samples; no safe-retry guidance for job creation (10). Microsoft publishes Online Services SLAs, but we couldn't open the document to confirm the Azure OpenAI clause, so part credit (5). Standard and global training are documented without a preview label, while the developer tier needs a preview api-version and GPT-5 RFT is by invitation (5).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-reliability):

Hosted APIs, MCP servers, models and platforms.

- 20, a public status page with component history (Statuspage, Instatus, BetterStack or the vendor's own).
- 0 to 30, the incident record for the last 90 days on that page. 30 for a clean record or trivial incidents only, 20 for minor incidents only, 10 for one major outage (an hour or more of a core API down, or errors across the board), 0 for several. 5 when there's no history we could read, and the note says so.
- 15, rate limits documented with numbers.
- 15, documented 429 or overload handling (Retry-After, backoff guidance), and idempotency keys or safe-retry guidance where writes are involved.
- 10, an SLA published for any paid tier.
- 10, the surface agents use is generally available, not beta or preview.

Local packages, SDKs, frameworks and stdio MCP servers.

- 20, installs from an official package with supported runtimes stated.
- 25, a public CI and test suite, passing on the default branch.
- 0 to 25, open crash or regression issues relative to activity (25 for few and handled, 0 for many, old and unanswered).
- 15, semver discipline and breaking changes called out in a changelog.
- 15, version 1.0 or later, or declared stable.

Protocols are read from their reference implementations, the public facilitators or servers, spec stability and test vectors.

## 4. Schema & documentation, 67 out of 100, up to 5.4 more on the total

Why it scored 67: A REST reference on Learn for the v1 data plane and the management plane; we didn't open a spec file this run (15). No llms.txt checked; the Markdown sources are public in the MicrosoftDocs/azure-ai-docs repository (5). The how-to guides say when to use SFT, DPO or RFT and list which models take which (15). The job body is the OpenAI shape, typed by method with hyperparameters per method (12). Python, REST and portal examples; no fine-tuning error reference found (8). Versioned by api-version and /openai/v1, but the Azure OpenAI 'what's new' page's newest dated section is May 2026 (12).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-schema):

APIs and MCP servers.

- 25, a machine-readable contract (a public OpenAPI file or similar; for MCP, typed JSON Schema inputs on every tool).
- 10, llms.txt or Markdown docs served for agents.
- 0 to 20, descriptions that say what a tool is for, when to use it and when not to, read from the tool definitions in the source or the API reference.
- 0 to 15, typed inputs with enums, constraints and required fields, and no free-form JSON blobs.
- 0 to 15, examples and documented error responses.
- 15, versioning and a public changelog.

Models are read from the API reference, the OpenAPI file, llms.txt, the structured-output and tool-use docs and the model cards. Frameworks from docs a model can follow, typed interfaces, examples and the API reference.

## 5. Maintenance & community, 55 out of 100, up to 3.9 more on the total

Why it scored 55: Foundry's 'what's new' for August 2026 was published on 1 September, and the docs carry fine-tuning retirement dates; no API release dated in the last 30 days found (20). One dated changelog section in the window; the Azure OpenAI page's last dated section is May 2026 (10). Microsoft Q&A and paid support exist, and the changelog is stale (10). The openai SDKs the docs point to are maintained by OpenAI; versions not rechecked this run (10). Package health not checked (5).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-maintenance):

- 0 to 30, time since the last release, or the last published model or API change for a closed service. 30 within 30 days, 20 within 90, 10 within 180, 0 older.
- 20, at least three releases or dated changelog entries in the last 90 days.
- 0 to 25, responsiveness. Issues and pull requests answered on GitHub (the open issues and how recent the replies are). For closed services, a public changelog and a support or community channel that answers, 0 to 15.
- 15, presence in the official MCP registry under a verified namespace (MCP servers), or current official SDKs (APIs and models).
- 10, package health, current dependencies and CI.

Models are read for deprecation notice periods and model churn rather than release counts.

## 6. Security & auth, 85 out of 100, up to 2.6 more on the total

Why it scored 85: Microsoft Entra ID tokens with Azure RBAC, or two rotatable resource keys in the `api-key` header; the key gives full data-plane access to the resource (30). Training needs Foundry User and deploying needs Foundry Owner, so the roles split the two (15). Returns job state and your own model's output, no third-party content (10). Azure Monitor resource logs, once a diagnostic setting is created, and the subscription activity log, for every customer (15). Microsoft's coordinated vulnerability disclosure and cloud bounty programmes (up to $100,000) and a SOC 2 Type 2 attestation for Azure; microsoft.com's security.txt expired on 2026-09-23 and we didn't check the advisory feed (15).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-security):

- 0 to 30, the credential model. 30 for OAuth 2.1 with scopes, or scoped and revocable keys with rotation. 20 for plain revocable API keys. 10 for one all-powerful key. 10 off when a secret can travel in a URL query string as a documented option.
- 0 to 20, read-only or least-privilege modes, and confirmation or approval for destructive actions.
- 0 to 15, prompt-injection posture where the tool returns untrusted content (documented mitigations or guidance). A tool that returns no untrusted content gets 10.
- 0 to 15, audit logs or per-call visibility for the operator.
- 0 to 20, a security programme. security.txt or a disclosure policy, a bug bounty, SOC 2 or ISO 27001, advisories handled in public.

Models are read for retention, whether API data trains models (and whether that's off by default), zero-retention options and certifications. Frameworks for telemetry defaults, approval hooks, guardrails and sandboxing.

## 7. Transparency & trust, 88 out of 100, up to 1.1 more on the total

Made of editorial 81, provenance 95.

Why it scored 88: Closed service under the Microsoft Product Terms, whose generative AI clause rules out training foundation models on Customer Data, and the openai SDK is Apache-2.0 (20). The data privacy page agrees with the terms. Training files and tuned models stay in the resource's geography, are encrypted with AES-256 or a customer key, can be deleted at any time and are exclusive to the customer; abuse-monitoring retention for fine-tuning isn't stated (26). A written retirement policy of at least 18 months after GA and 60 days' notice, with separate training and deployment retirement dates for each tunable model, readable through the Models API (20). Data locations are stated per deployment type, including that global training may process in any geography; the subprocessor list wasn't checked this run (15).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-transparency):

- 0 to 30, source availability and licence clarity. 30 for open source under an OSI licence, 15 for closed with clear terms, 0 for unclear terms.
- 0 to 30, data handling and retention statements that agree with each other (privacy policy, DPA, retention periods, subprocessors).
- 0 to 20, a deprecation policy or notices with dates.
- 0 to 20, telemetry disclosed with an opt-out (local software), or subprocessors and data locations disclosed (hosted).

The other half of Transparency and trust is the provenance score, computed from checked facts (below). The category score is the mean of the two.

Provenance checks not met in full (half of this category, computed from checked facts):

- security.txt: published but past its Expires date (5 of 10)

## What we couldn't check

What we couldn't read counted as absent. Publishing it on a page a plain HTTP fetch can read (not only in a browser) lets the next check count it.

- We couldn't open the Online Services SLA document to confirm the Azure OpenAI uptime commitment.
- The pricing page's fine-tuning table didn't render; prices here come from the Retail Prices API for gpt-4.1 models in Sweden Central, and the open-model and RFT meters weren't queried.
- The listing's old pricing note quoted $2 per 1M training tokens for GPT-4.1 global; the Retail Prices API says $25, and $2 is the tuned model's input price, so we corrected it.
- Whether failed or cancelled jobs are billed isn't stated in the pages we read.

## Weaknesses

- No weight export; checkpoints copy only between Azure resources
- $1.70 an hour hosting on Standard deployments, and deletion after 15 idle days
- GPT-4.1 training at $25 per 1M tokens globally, and no free tier without a card
- Deployment goes through management.azure.com with a separate credential and the Foundry Owner role
- The Azure OpenAI 'what's new' page hasn't had a dated section since May 2026

## What costs an agent a turn today

The notes we give agents before they call it. Each one is a workaround an agent shouldn't need.

- Point the OpenAI SDK at https://<resource>.openai.azure.com/openai/v1 with the `api-key` header or an Entra token; job, file and checkpoint calls are the OpenAI shapes
- Read prices from the Azure Retail Prices API (meters named like 'gpt-4.1 FT Training global'), not the pricing page, which needs a browser
- Keep at most 3 jobs running and 20 queued per resource, and keep training files under 512 MB and 1 GB in total
- Create the deployment through the Resource Manager API with a Foundry Owner identity, then call it at least once a fortnight or it's deleted
- Query the Models API for `deprecationDate` before choosing a base model

## What the review panel asked for

- current dated release notes
- State failed-job billing
- Make pricing table readable

## When it's done

Send what changed and where it's published as a dispute (https://www.anchorterminal.com/builders/#disputes, or `POST https://www.anchorterminal.com/api/v1/contact` with `"kind": "dispute"`). Disputes are answered in public, and the listing is checked again by the same checklist. Paying for an audit or a listing claim changes nothing here.