OpenAI 'model does not exist' Error (model_not_found): Causes & Fix

OpenAI returns an HTTP 404 with code model_not_found when the model string in your request doesn’t resolve to something your account can call. The message is deliberately ambiguous — “does not exist or you do not have access” — because OpenAI won’t reveal whether a model exists to accounts that can’t use it. So the fix is a short checklist: confirm the name, your access, and the endpoint.

What this error means

{
"error": {
  "message": "The model 'gpt-5-turbo' does not exist or you do not have access to it.",
  "type": "invalid_request_error",
  "param": null,
  "code": "model_not_found"
}
}

It bundles several distinct situations under one message:

Common causes

• Typo or wrong version. gpt-4o vs gpt-4-o, or a dated snapshot (-2024-08-06) that’s wrong. Model ids are exact strings.
• Retired or renamed model. A model you used months ago was deprecated and shut down — see the deprecations tracker for what replaced it.
• No access on your account. Newer or higher-end models can require a verified organization or a paid account; unverified accounts get the same 404.
• Wrong endpoint for the model type. Embeddings models don’t work on /chat/completions; chat models don’t work on the legacy /completions. Mismatched pairs surface as model_not_found.
• Wrong project/org key. A key scoped to a project without access to that model.
• Fine-tuned model id is off. Fine-tunes have long ids (ft:gpt-4o-mini:org::id); a wrong suffix 404s.

How to fix it

List the models you can actually call

The fastest disambiguation: ask the API what your key has access to.

curl https://api.openai.com/v1/models \
-H "Authorization: Bearer $OPENAI_API_KEY" | python3 -c "import sys,json;[print(m['id']) for m in json.load(sys.stdin)['data']]"

from openai import OpenAI
client = OpenAI()

available = {m.id for m in client.models.list().data}
target = "gpt-4o"
if target not in available:
  raise SystemExit(f"{target!r} not available to this key. "
                   f"Did you mean one of: {sorted(a for a in available if 'gpt-4o' in a)}")
print("OK:", target)

If the model is real but you lack access

• Verify your organization in the dashboard (Settings → Organization) — some models gate on verification.
• Add billing / move off the free tier if the model requires a paid account.
• Confirm the key’s project has the model enabled.

If the model was retired

Check the model deprecations tracker for the official replacement and update your model string. Pin to a dated snapshot you control so a future rename doesn’t silently break you — and handle the 404 gracefully with a fallback:

import openai
from openai import OpenAI
client = OpenAI()

PREFERRED = "gpt-5.5"
FALLBACK = "gpt-4o"

def create(messages, model=PREFERRED):
  try:
      return client.chat.completions.create(model=model, messages=messages)
  except openai.NotFoundError:
      # Model id invalid/retired/no-access — fall back and log to fix the config.
      return client.chat.completions.create(model=FALLBACK, messages=messages)

How to prevent it

• Centralize model ids in config, not scattered string literals, so updates are one-line.
• Validate the id against models.list() at startup and fail fast with a clear message.
• Watch deprecations so a shutdown doesn’t surprise you — the deprecations tracker lists retirement dates and replacements.
• Match the endpoint to the model type (chat vs embeddings vs responses).

Related errors

• A 401 is authentication, not the model — see invalid authentication.
• Rate/quota issues are separate: 429 rate limit.

What to do next

1. List your available models and copy the exact id.
2. Check the deprecations tracker if a once-working model now 404s.
3. Verify your org / billing if the model is real but inaccessible.
4. Add a fallback so a bad id degrades gracefully instead of crashing.