OpenAI API breaking changes (2026): complete tracker and fixes
openai 7 breaking / deprecation entries 2026
OpenAI ships API and SDK changes continuously. Most are additive — but the ones that are not can silently break integrations that worked yesterday. This page is the complete, continuously regenerated record of every OpenAI breaking change and deprecation in the mendapi change database: 7 entries (7 breaking, 0 deprecation), each linked to its upstream source and, where possible, to a concrete fix.
How to read this page
Every entry carries a fixability verdict from the mendapi adjudication pipeline: code-fixable means a mechanical code change repairs the break (3 entries); not code-fixable means the change requires a business decision, credential provisioning, or upstream action that no codemod can produce (4 entries). Entries marked auto-fix pack have a deterministic, gold-tested migration pack that mendapi fix can apply locally.
Am I affected?
Reading a changelog tells you what changed — not whether your code uses the affected surface. The mendapi scanner answers that locally, in about 30 seconds, without any code leaving your machine:
npx mendapi sync # -> one network call: pulls the change feed into a local database npx mendapi scan # -> N findings: file, line, provider, confidence mendapi deps --match # -> which of the changes below actually hit your endpoints
See Getting Started for the full workflow and the security model for why zero code upload is the default and only mode.
Complete OpenAI breaking change timeline
| Date | Change | Ref |
|---|---|---|
| 2026-07-31 | request-media-type-removed: POST /skills/{skill_id} application/x-www-form-urlencoded breaking code-fixable |
spec diff |
| 2026-07-31 | request-media-type-removed: POST /responses/input_tokens application/x-www-form-urlencoded breaking code-fixable |
spec diff |
| 2026-07-31 | request-media-type-removed: POST /responses/compact application/x-www-form-urlencoded breaking code-fixable |
spec diff |
| 2026-07-31 | v7.3.0 breaking not code-fixable openai-node v7.3.0: features are additive (content provenance checks) |
source |
| 2026-07-27 | v7.0.0 breaking not code-fixable openai-node v7.0.0: the only breaking item is 'require Node.js 22 and codify version support' (#2026) |
source |
| 2026-07-17 | v2.46.0 breaking not code-fixable Additive endpoints and compatibility-preserving fixes |
source |
| 2026-06-01 | v6.40.0 breaking not code-fixable ActionSearch.query loosened to optional (non-breaking for consumers) and migrate CLI removed (tooling) |
source |
Automated migration packs for OpenAI
mendapi ships 1 deterministic migration pack for OpenAI — verified, idempotent codemods you can apply locally with mendapi fix. Full details in the pack catalog.
openai-v3-to-v4— openai-node v3 -> v4 SDK migration (upstream reference)
Recent OpenAI releases adjudicated non-breaking
Not paging you is the other half of the job. The same pipeline that produced the timeline above also cleared 27 OpenAI changes as additive or fix-only — releases an integrator can upgrade through without a migration. The most recent:
| Date | Release | Ref |
|---|---|---|
| 2026-08-03 | v7.4.0 additive api: Add gpt-5.5 model and tool metadata fields |
source |
| 2026-08-03 | v2.52.1 fix |
source |
| 2026-08-03 | v2.53.0 additive api: Add gpt-5.5 and tool name/namespace to Responses types |
source |
| 2026-07-31 | v2.52.0 additive api: content provenance checks |
source |
| 2026-07-30 | v7.2.0 additive api: fast tier |
source |
| 2026-07-30 | v2.51.0 additive api: fast tier |
source |
| 2026-07-28 | v7.1.0 additive api: transcription model updates |
source |
| 2026-07-28 | v2.50.0 additive api: transcription model updates |
source |
Measured: two feeds, and they disagree about what breaking means
The OpenAI entries above are the one set in this tracker series fed by both pipelines: an OpenAPI spec-diff corridor over OpenAI's published spec, and the openai-node / openai-python release feeds. The two feeds split cleanly, and the split is the useful finding. Of the 7 breaking and deprecation entries above, 3 are code-fixable and 4 are not — and the verdict tracks the feed, not the severity of the headline.
The spec corridor: quiet, and every break is mechanical. Diffing the published spec across the May-to-July 2026 corridor produced 821 raw diff records and exactly 3 breaking changes — all of them the same change: application/x-www-form-urlencoded request bodies dropped from three endpoints. All 3 were adjudicated code-fixable: the endpoints survive accepting JSON, so the repair is a mechanical transport-encoding swap. The corridor baseline is re-verified record-for-record against live diff runs (2 pairs, 2 stable, 0 regressions), so this section cannot silently drift from the evidence it cites.
The release feed: loud, and nothing in it is patchable. All 4 breaking flags captured from SDK release notes were adjudicated not-code-fixable — runtime floor bumps, distribution-channel removals, tooling churn. “v7.0.0” is the pattern case: openai-node v7.0.0: the only breaking item is 'require Node.js 22 and codify version support' — a runtime upgrade decision, not a consumer code migration. A team paging on release-note keywords does 4 migrations' worth of triage here and ships zero code changes; the wire-contract breaks that did need patching came from the corridor the release notes never mentioned. Measuring both feeds, and adjudicating each entry, is how the two stop being confused.
Monitoring OpenAI going forward
This page is regenerated from the change database, which is fed by release feeds, changelog capture and OpenAPI spec diffing — so it does not depend on OpenAI writing a perfect changelog. mendapi monitors 84125 change records across 20 providers; hosted plans watch your repos continuously and open fix PRs when a change lands. See pricing.
Frequently asked questions
How do I check if my codebase is affected by OpenAI breaking changes?
Run npx mendapi scan in your repo. It statically matches your files against the 7 OpenAI breaking changes and deprecations in this tracker and reports file, line and confidence for every hit — locally, in about 30 seconds, with zero network calls.
How many OpenAI API breaking changes are on record?
This tracker records 7 entries for OpenAI (7 breaking changes and 0 deprecations, 2026). Each entry is adjudicated: 3 are code-fixable, 4 require a decision no codemod can make.
Are there automated fixes for OpenAI breaking changes?
Yes — mendapi ships 1 deterministic migration pack for OpenAI. mendapi fix applies them locally as reviewable diffs; every pack is gold-tested and idempotent.
Does my code get uploaded when scanning for OpenAI changes?
No. mendapi scan, deps and fix run with zero network code, mechanically enforced by the test suite. Optional hosted reporting is metadata-only, secrets-redacted, and requires an explicit flag.
Related
- openai-node v3 to v4 migration — the full migration guide with pack-backed fixes.
- 821 OpenAI spec changes, 3 that can break a client: the honest negative — the long-form methodology report behind this provider's tracked changes.
- OpenAI drop-in replacement APIs — what a compatible base URL does and does not shield you from.
- API breaking change detection — how upstream breakage is caught before it ships, and why default severity gates stay silent.
- All breaking-change reports — per-change deep dives with migration guides.
- All monitored providers — live change counts.
- Migration pack catalog — every deterministic fix mendapi ships.