Add GPT-5 to Code Eval #180
Conversation
WalkthroughThe UI title/header was updated. The default fallback for model1 changed to “Claude Opus 4.1.” The model catalog replaced “Claude Sonnet 4” and “GPT-4.1” with “Claude Opus 4.1” and “GPT-5.” Response routing now sends “GPT” models to OpenAI (OPENAI_API_KEY, max_completion_tokens) and others to OpenRouter (OPENROUTER_API_KEY, max_tokens). Changes
Sequence Diagram(s)sequenceDiagram
actor User
participant App as App UI
participant MS as ModelService
participant OA as OpenAI API
participant OR as OpenRouter API
User->>App: Select models / submit prompt
App->>MS: get_model_response_async(model_name, prompt)
alt model_name contains "GPT"
MS->>OA: acompletion(api_key=OPENAI_API_KEY, max_completion_tokens=2000)
OA-->>MS: response stream
else
MS->>OR: acompletion(api_key=OPENROUTER_API_KEY, max_tokens=2000)
OR-->>MS: response stream
end
MS-->>App: stream tokens / final text
App-->>User: display comparison output
Estimated code review effort🎯 2 (Simple) | ⏱️ ~8 minutes Possibly related PRs
Poem
✨ Finishing Touches
🧪 Generate unit tests
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. 🪧 TipsChatThere are 3 ways to chat with CodeRabbit:
SupportNeed help? Create a ticket on our support page for assistance with any issues or questions. CodeRabbit Commands (Invoked using PR comments)
Other keywords and placeholders
CodeRabbit Configuration File (
|
![coderabbitai[bot]](https://cdn.statically.io/img/avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
There was a problem hiding this comment.
Actionable comments posted: 2
🔭 Outside diff range comments (2)
code-model-comparison/model_service.py (1)
75-84: Reflect correct API key environment variable in error messages
Adjust the exception handler incode-model-comparison/model_service.py(around lines 75–84) so that invalid/missing API key errors referenceOPENAI_API_KEYwhen routing to OpenAI andOPENROUTER_API_KEYotherwise.• File: code-model-comparison/model_service.py
Lines: 75–84except Exception as e: error_msg = f"Error generating response: {str(e)}" - if "api_key" in str(e).lower() or "authentication" in str(e).lower(): - error_msg = "Error: Invalid or missing API key. Please check your OPENROUTER_API_KEY configuration." + if "api_key" in str(e).lower() or "authentication" in str(e).lower(): + # Dynamically reference the correct env var based on model routing + env_var = "OPENAI_API_KEY" if "GPT" in model_name else "OPENROUTER_API_KEY" + error_msg = ( + f"Error: Invalid or missing API key. " + f"Please check your {env_var} configuration." + ) elif "quota" in str(e).lower() or "limit" in str(e).lower(): error_msg = "Error: API quota exceeded or rate limit reached. Please try again later." elif "model" in str(e).lower(): error_msg = f"Error: Model '{model_name}' is not available or has issues. Please try a different model."code-model-comparison/app.py (1)
65-71: Broken logo URL (double protocol) in “Powered by” strip.The second image src has a duplicated https://, so the Opik logo won’t render.
- <img src="https://files.buildwithfern.com/https://opik.docs.buildwithfern.com/docs/opik/2025-08-01T07:08:31.326Z/img/logo-dark-mode.svg" width="100"> + <img src="https://files.buildwithfern.com/opik.docs.buildwithfern.com/docs/opik/2025-08-01T07:08:31.326Z/img/logo-dark-mode.svg" width="100">
🧹 Nitpick comments (1)
code-model-comparison/app.py (1)
91-94: Unreachable fallback branch; simplify default selection.Because you stop execution when all_models is empty (Lines 82–85), the else "Claude Opus 4.1" fallback on Line 92 is unreachable. Simplify to avoid config drift.
-if default_model1 not in all_models: - default_model1 = all_models[0] if all_models else "Claude Opus 4.1" +if default_model1 not in all_models: + default_model1 = all_models[0]
📜 Review details
Configuration used: CodeRabbit UI
Review profile: CHILL
Plan: Pro
📒 Files selected for processing (2)
code-model-comparison/app.py(3 hunks)code-model-comparison/model_service.py(2 hunks)
🔇 Additional comments (2)
code-model-comparison/app.py (2)
13-13: Branding/title change LGTM.
62-62: Main header change LGTM.
| "Claude Opus 4.1": "openrouter/anthropic/claude-opus-4.1", | ||
| "Qwen3-Coder": "openrouter/qwen/qwen3-coder", | ||
| "Gemini 2.5 Flash": "openrouter/google/gemini-2.5-flash", | ||
| "GPT-4.1": "openrouter/openai/gpt-4.1", | ||
| "GPT-5": "gpt-5", | ||
| } |
There was a problem hiding this comment.
💡 Verification agent
🧩 Analysis chain
Model catalog updates look consistent, but verify identifiers exist upstream.
"Claude Opus 4.1" and "GPT-5" entries look fine. Ensure LiteLLM recognizes these identifiers and providers as configured.
Run web checks to confirm model IDs and parameter support:
If needed, I can adjust the mapping to provider-qualified IDs (e.g., openai/gpt-5) or add a fallback route.
🌐 Web query:
Does LiteLLM support the model id "gpt-5" with acompletion, and should we pass max_tokens or max_completion_tokens for streaming?
💡 Result:
Short answer: Yes — LiteLLM supports GPT‑5 model IDs (e.g. "openai/gpt-5" or "azure/gpt-5") for its completion APIs, and for GPT‑5 you should use max_completion_tokens to control the streamed completion length (LiteLLM will also accept max_tokens and map it to max_completion_tokens for GPT‑5, but passing max_completion_tokens is clearer/explicit). (docs.litellm.ai)
Quick notes and example:
- Supported IDs: LiteLLM lists OpenAI/Azure GPT‑5 model IDs such as openai/gpt-5 and openai/gpt-5-chat (and provider-prefixed variants). (docs.litellm.ai)
- Params: LiteLLM exposes both max_tokens and max_completion_tokens; for GPT‑5 it maps max_tokens → max_completion_tokens, and you can (and probably should) pass max_completion_tokens when streaming so you explicitly set the completion token cap. (docs.litellm.ai)
Example (Python):
from litellm import completion
resp = completion(
model="openai/gpt-5",
messages=[{"role":"user","content":"Write a short summary of X."}],
stream=True,
max_completion_tokens=250
)
for chunk in resp:
print(chunk.choices[0].delta)
If you want, tell me which provider/model string you’re using (openai/…, azure/…, or bare "gpt-5") and I’ll give a tailored snippet and confirm the exact param mapping for your LiteLLM version.
Citations:
- 1: https://docs.litellm.ai/release_notes/v1-75-5?utm_source=chatgpt.com
- 2: https://docs.litellm.ai/release_notes/v1-75-5?utm_source=chatgpt.com
- 3: https://docs.litellm.ai/docs/completion/input?utm_source=chatgpt.com
🌐 Web query:
Is "openrouter/anthropic/claude-opus-4.1" a valid model identifier on OpenRouter via LiteLLM?
💡 Result:
Short answer: No. The valid Anthropic Opus identifier on OpenRouter via LiteLLM uses the Claude 3 naming, e.g., openrouter/anthropic/claude-3-opus-20240229, not “claude-opus-4.1.”[1][3]
Details:
- LiteLLM’s OpenRouter convention is openrouter/provider/model-name, and Anthropic’s Opus listing on OpenRouter uses the Claude 3 format such as claude-3-opus-20240229.[1]
- LiteLLM docs state you can call any OpenRouter model by passing model="openrouter/", with examples including openrouter/anthropic/claude-2 and the Claude 3 variants, but there is no “claude-opus-4.1” pattern shown.[3]
- A third-party page shows a string like openrouter/anthropic/claude-opus-4.1, but this appears inconsistent with OpenRouter’s established Claude 3 model identifiers and should not be relied upon without confirmation on the OpenRouter models list.[2]
Recommendation:
- Use the confirmed identifier: openrouter/anthropic/claude-3-opus-20240229.[1]
- If you need the latest Sonnet/Haiku/Opus variant, check the OpenRouter model list and plug it into LiteLLM as openrouter/anthropic/.[3]
Citations:
- 1: https://saptak.in/writing/2025/05/10/google-adk-masterclass-part3
- 2: https://lightning.ai/lightning-purchase-test/studios/compare-openai-gpt-5-and-claude-opus-4-1-for-code-generation
- 3: https://docs.litellm.ai/docs/providers/openrouter
- 4: https://docs.litellm.ai/docs/completion/input
- 5: https://www.bitdoze.com/google-adk-openrouter-models/
- 6: https://python.useinstructor.com/integrations/openrouter/
- 7: https://github.com/BerriAI/litellm/blob/main/README.md
- 8: https://denshub.com/en/choosing-llm-gateway/
- 9: https://www.truefoundry.com/blog/litellm-vs-openrouter
- 10: https://ai.plainenglish.io/power-up-your-ai-agents-unlocking-a-universe-of-models-with-litellm-and-openrouter-7c25262cbbe9
Fix invalid LiteLLM model identifiers and use explicit max_completion_tokens
Please update the model mapping in code-model-comparison/model_service.py and streaming calls as follows:
-
In the AVAILABLE_MODELS mapping (lines 9–13):
- "Claude Opus 4.1": "openrouter/anthropic/claude-opus-4.1", + "Claude Opus 4.1": "openrouter/anthropic/claude-3-opus-20240229", # valid OpenRouter Anthropic Opus ID - "GPT-5": "gpt-5", + "GPT-5": "openai/gpt-5", # provider-prefixed for LiteLLM compatibility
-
In your GPT-5 streaming completion branches (around lines 49–64):
• Replace any usage ofmax_tokenswithmax_completion_tokensto explicitly set the streamed token cap (LiteLLM will mapmax_tokens→max_completion_tokensbut explicit is clearer).resp = completion( model="openai/gpt-5", messages=…, stream=True, - max_tokens=250 + max_completion_tokens=250 )
These changes ensure LiteLLM recognizes both model IDs and correctly applies the streaming token limit.
🤖 Prompt for AI Agents
In code-model-comparison/model_service.py lines 9–13 and the GPT-5 streaming
branches around lines 49–64, update the AVAILABLE_MODELS mapping so the GPT-5
entry uses the valid LiteLLM identifier ("openai/gpt-5") instead of "gpt-5", and
in all GPT-5 stream calls replace max_tokens with max_completion_tokens (e.g.,
max_completion_tokens=250) so the streamed token cap is set explicitly; keep the
rest of the call arguments the same.
| if "GPT" in model_name: | ||
| response = await acompletion( | ||
| model=model_mapping, | ||
| messages=messages, | ||
| api_key=os.getenv("OPENAI_API_KEY"), | ||
| max_completion_tokens=2000, | ||
| stream=True, | ||
| ) | ||
| else: | ||
| response = await acompletion( | ||
| model=model_mapping, | ||
| messages=messages, | ||
| api_key=os.getenv("OPENROUTER_API_KEY"), | ||
| max_tokens=2000, | ||
| stream=True, | ||
| ) |
There was a problem hiding this comment.
Use max_tokens for LiteLLM acompletion; align params across routes.
The GPT branch passes max_completion_tokens, which acompletion may not accept. Use max_tokens in both branches to avoid parameter mismatch and parity issues.
Apply this minimal diff:
- if "GPT" in model_name:
- response = await acompletion(
- model=model_mapping,
- messages=messages,
- api_key=os.getenv("OPENAI_API_KEY"),
- max_completion_tokens=2000,
- stream=True,
- )
+ if "GPT" in model_name:
+ response = await acompletion(
+ model=model_mapping,
+ messages=messages,
+ api_key=os.getenv("OPENAI_API_KEY"),
+ max_tokens=2000,
+ stream=True,
+ )Optionally, dedupe branches by deriving api_key and calling acompletion once:
- if "GPT" in model_name:
- response = await acompletion(
- model=model_mapping,
- messages=messages,
- api_key=os.getenv("OPENAI_API_KEY"),
- max_tokens=2000,
- stream=True,
- )
- else:
- response = await acompletion(
- model=model_mapping,
- messages=messages,
- api_key=os.getenv("OPENROUTER_API_KEY"),
- max_tokens=2000,
- stream=True,
- )
+ is_gpt = "GPT" in model_name
+ response = await acompletion(
+ model=model_mapping,
+ messages=messages,
+ api_key=os.getenv("OPENAI_API_KEY") if is_gpt else os.getenv("OPENROUTER_API_KEY"),
+ max_tokens=2000,
+ stream=True,
+ )📝 Committable suggestion
‼️ IMPORTANT
Carefully review the code before committing. Ensure that it accurately replaces the highlighted code, contains no missing lines, and has no issues with indentation. Thoroughly test & benchmark the code to ensure it meets the requirements.
| if "GPT" in model_name: | |
| response = await acompletion( | |
| model=model_mapping, | |
| messages=messages, | |
| api_key=os.getenv("OPENAI_API_KEY"), | |
| max_completion_tokens=2000, | |
| stream=True, | |
| ) | |
| else: | |
| response = await acompletion( | |
| model=model_mapping, | |
| messages=messages, | |
| api_key=os.getenv("OPENROUTER_API_KEY"), | |
| max_tokens=2000, | |
| stream=True, | |
| ) | |
| if "GPT" in model_name: | |
| response = await acompletion( | |
| model=model_mapping, | |
| messages=messages, | |
| api_key=os.getenv("OPENAI_API_KEY"), | |
| max_tokens=2000, | |
| stream=True, | |
| ) | |
| else: | |
| response = await acompletion( | |
| model=model_mapping, | |
| messages=messages, | |
| api_key=os.getenv("OPENROUTER_API_KEY"), | |
| max_tokens=2000, | |
| stream=True, | |
| ) |
🤖 Prompt for AI Agents
In code-model-comparison/model_service.py around lines 49 to 64, the GPT branch
passes max_completion_tokens while the other branch uses max_tokens, causing
acompletion parameter mismatch; change the GPT branch to use max_tokens=2000
(matching the other branch) and align other parameter names, and optionally
simplify by setting api_key = os.getenv("OPENAI_API_KEY") if "GPT" in model_name
else os.getenv("OPENROUTER_API_KEY") and call acompletion once with the derived
api_key and shared args (model=model_mapping, messages=messages,
api_key=api_key, max_tokens=2000, stream=True).
1 participant
Summary by CodeRabbit
New Features
Refactor
Style