Update LLM config to support OpenAI 5-series and Gemini 3-series

[KianBahasadri]

The current list of compatible LLMs in l2p/llm/utils/llm.yaml is outdated and missing support for the latest state-of-the-art models.

I would like to see support added for the following:

New Models

OpenAI (5-series)

• gpt-5.2: General purpose flagship (released Dec 2025).
• gpt-5.2-pro: High-reasoning variant with configurable effort.
• gpt-5.3-codex: Specialized agentic coding model (released Feb 2026).

Google (Gemini 3-series)

• gemini-3.0-pro: Multimodal model with massive context window.
• gemini-3.0-flash: Low-latency variant.

Proposal

I am happy to submit a PR for this.

Quick question regarding the implementation: Is it alright if I copy the same model_params (e.g. temperature, top_p) from the existing configurations as defaults for these new models, or should I use different values?

[haz]

PR away!

Check the API to see what settings are required. Anything that overlaps should be ok to copy the values but @MarcusTantakoun would know better

[MarcusTantakoun]

Sounds good to me! I agree that it could use an update (seems like there are new GPT models coming out everyday...)

Touching on that, all the LLM classes takes a config path argument for a .yaml file which contains the LLM specs. You can simply just pass your own YAML file with your collection of LLMs if you need to :) -- you should probably do this anyways if you want to mess around with the model parameters.

class OPENAI(BaseLLM):
    def __init__(
        self,
        model: str,
        config_path: str = "l2p/llm/utils/llm.yaml",
        provider: str = "openai",
        api_key: str | None = None,
        base_url: str = "https://api.openai.com/v1/",
    ) -> None:

https://github.com/AI-Planning/l2p/blob/main/l2p/llm/openai.py#L20

For the PR, I would keep the model params the same -- just change the API cost/context window accordingly to the respective model you add.