feat(openai): add new models [bot]

[models-bot]

Auto-generated by model-addition-agent for provider openai.

---

Note

Low Risk
Low risk: adds new OpenAI model metadata YAMLs only (no runtime code changes), mainly affecting model availability and cost/limit configuration.

Overview
Adds three new OpenAI model specification files for realtime usage: gpt-realtime-2, gpt-realtime-translate, and gpt-realtime-whisper.

Each model entry defines pricing, token/second limits, supported modalities/modes, and any parameter overrides (e.g., max_tokens for translate, removeParams for whisper), enabling them to be selected/configured by the existing provider model catalog.

^{Reviewed by Cursor Bugbot for commit 728ac78. Bugbot is set up for automated code reviews on this repo. Configure here.}

[github-actions]

/test-models

[harshiv-26]

Gateway test results

• Total: 3
• Passed: 0
• Failed: 1
• Validation failed: 0
• Errored: 0
• Skipped: 2
• Success rate: 0.0%

Provider	Model	Scenarios
openai	gpt-realtime-2	skipped: skip-check
openai	gpt-realtime-translate	skipped: skip-check
openai	gpt-realtime-whisper	failure: params

Failures (1)

openai/gpt-realtime-whisper — params (failure)

Error:

Traceback (most recent call last):
  File "/tmp/tmph2iiuzi1/snippet.py", line 9, in <module>
    response = client.audio.transcriptions.create(
               ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.11/site-packages/openai/_utils/_utils.py", line 286, in wrapper
    return func(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.11/site-packages/openai/resources/audio/transcriptions.py", line 333, in create
    return self._post(  # type: ignore[return-value]
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.11/site-packages/openai/_base_client.py", line 1259, in post
    return cast(ResponseT, self.request(cast_to, opts, stream=stream, stream_cls=stream_cls))
                           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/local/lib/python3.11/site-packages/openai/_base_client.py", line 1047, in request
    raise self._make_status_error_from_response(err.response) from None
openai.NotFoundError: Error code: 404 - {'status': 'failure', 'message': 'openai error: Invalid URL (POST /v1/audio/transcriptions)', 'error': {'message': 'openai error: Invalid URL (POST /v1/audio/transcriptions)', 'type': 'invalid_request_error', 'code': '404'}, 'provider': 'openai'}

Code snippet

# OpenAI-compatible STT (works with OpenAI and Groq)
from pathlib import Path
from openai import OpenAI

client = OpenAI(api_key="***", base_url="https://internal.devtest.truefoundry.tech/api/llm/api/llm")

audio_path = Path(__file__).parent / "hello.wav"
with open(audio_path, "rb") as audio_file:
    response = client.audio.transcriptions.create(
        model="test-v2-openai/gpt-realtime-whisper",
        file=audio_file,
    )

_stt_output = response.text
print(_stt_output)

assert _stt_output, "VALIDATION FAILED: Transcription returned empty text"
assert len(_stt_output.strip()) > 0, f"VALIDATION FAILED: Transcription is blank: '{_stt_output}'"

Skipped (2)

openai/gpt-realtime-2 — skip-check (skipped)

Skip reason:

unsupported mode 'realtime'

openai/gpt-realtime-translate — skip-check (skipped)

Skip reason:

unsupported mode 'realtime'

[cursor]

Thinking model missing removeParams and max_completion_tokens config

Medium Severity

gpt-realtime-2 sets thinking: true but lacks removeParams and a params entry for max_completion_tokens. Every other OpenAI thinking model (o1, o3, o4-mini, gpt-5, gpt-5-mini, gpt-5-nano, gpt-5-codex, gpt-5-pro, computer-use-preview) removes at least max_tokens and usually temperature/top_p via removeParams, and adds max_completion_tokens to params. Without these, the model inherits temperature, max_tokens, top_p, n, etc. from default.yaml, which thinking/reasoning models typically don't support.

 

^{Reviewed by Cursor Bugbot for commit 39abcc2. Configure here.}

[cursor]

Translation model missing removeParams for tool-related parameters

Medium Severity

gpt-realtime-translate has no features (no function calling) and no removeParams, so it inherits tool_choice and parallel_tool_calls from default.yaml. Every other OpenAI audio model without function calling — including gpt-realtime-whisper added in this same PR, plus whisper-1, gpt-4o-transcribe, gpt-4o-mini-transcribe, and gpt-4o-transcribe-diarize — removes these params via removeParams. This translation-only model shouldn't expose tool-related parameters.

 

^{Reviewed by Cursor Bugbot for commit 39abcc2. Configure here.}

[github-actions]

/test-models

[cursor]

Cursor Bugbot has reviewed your changes and found 1 potential issue.

There are 3 total unresolved issues (including 2 from previous reviews).

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit 728ac78. Configure here.}