Conversation
![cursor[bot]](https://avatars.githubusercontent.com/in/1210556?s=60&v=4){kind=small}
There was a problem hiding this comment.
Cursor Bugbot has reviewed your changes and found 4 potential issues.
Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.
| mode: chat | ||
| model: gpt-5.4-mini | ||
| params: | ||
| - defaultValue: 128000 |
There was a problem hiding this comment.
Default max_tokens set to maximum value in alias
Medium Severity
The defaultValue for max_tokens in gpt-5.4-mini.yaml is 128000, which equals the maxValue. The dated counterpart gpt-5.4-mini-2026-03-17.yaml uses defaultValue: 128. Since the non-dated alias resolves to the same underlying model, these defaults are inconsistent. A default equal to the maximum means every request consumes the full token budget by default, likely leading to unexpected cost and latency.
Additional Locations (1)
| minValue: 1 | ||
| sources: | ||
| - https://developers.openai.com/api/docs/models/gpt-5.4-nano | ||
| thinking: true |
There was a problem hiding this comment.
Nano alias missing reasoning_effort parameter despite thinking support
Medium Severity
gpt-5.4-nano.yaml has thinking: true but is missing the reasoning_effort parameter that its dated counterpart gpt-5.4-nano-2026-03-17.yaml includes. Users of the non-dated alias won't be able to control reasoning effort level, resulting in an inconsistent experience compared to the dated model name.
Additional Locations (1)
| - function_calling | ||
| - structured_output | ||
| - tool_choice | ||
| - system_messages |
There was a problem hiding this comment.
Nano alias missing prompt_caching despite having cache cost
Medium Severity
gpt-5.4-nano.yaml defines a cache_read_input_token_cost in its costs but omits prompt_caching from its features list. The dated counterpart gpt-5.4-nano-2026-03-17.yaml correctly includes prompt_caching. This inconsistency means the feature may not be advertised or enabled for users of the non-dated alias despite the cost being configured.
Additional Locations (1)
| - prompt_caching | ||
| - tools | ||
| - tool_choice | ||
| - system_messages |
There was a problem hiding this comment.
Mini alias and dated version have swapped features
Medium Severity
The features lists for gpt-5.4-mini-2026-03-17.yaml and gpt-5.4-mini.yaml are inconsistent in a contradictory way. The dated version lists parallel_function_calling but not tools, while the non-dated alias lists tools but not parallel_function_calling. Since the alias resolves to the same underlying model, both files represent the same capabilities yet advertise different feature sets. This could cause tool-calling behavior to differ depending on which model name a user specifies.
Additional Locations (1)
1 participant
Auto-generated by poc-agent for provider
openai.Note
Low Risk
Low risk metadata-only change: updates/adds OpenAI model configuration YAMLs (pricing, limits, feature flags) without touching runtime code paths.
Overview
Updates OpenAI provider model definitions for
gpt-5.4-miniandgpt-5.4-nano, and adds dated variants (*-2026-03-17).The YAMLs now include explicit token pricing, capability flags (e.g., tool/function calling, structured outputs, prompt caching), context/output limits (up to
400000context /128000output), supported modalities (text+image in, text out), and exposed params likemax_tokensandreasoning_effort, with updated documentation sources andthinking: true.Written by Cursor Bugbot for commit 5159fcf. This will update automatically on new commits. Configure here.