feat: add GPT-5 model family support #92

[bahdan111]

Applicable issues

• Connected to Core: Switch to GPT-5 series models #92 (Haven't updated default models and configs)

Description of changes

Adds GPT-5 model family support with conditional parameter handling:

• New models: GPT_5_MINI, GPT_5_1, GPT_5_2
• New enums: ReasoningEffortEnum, VerbosityEnum
• LLMModelConfig Pydantic validator enforces model-family constraints:
  • GPT-5: rejects seed, requires reasoning_effort, validates temperature=1 when reasoning is enabled
  • Non-GPT-5: rejects reasoning_effort and verbosity
• Refactored get_chat_model() - moved params validation into LLMModelConfig
• Removed unused **kwargs from get_chat_model() and get_embeddings_model()
• Moved reasoning_effort and verbosity defaults out of LangChainSettings into LLMModelConfig
• Default model remains GPT-4.1
• Fix bug with default embedding model incorrectly set in channel config, correct assignment: llm_model: Mapped[str] = mapped_column( default=langchain_settings.embedding_default_model.value )

Checklist

• Title of the pull request follows Conventional Commits specification
• Deployed and tested in a Review environment.

By submitting this pull request, I confirm that my contribution is made under the terms of the MIT license.

[bahdan111]

/deploy-review

GitHub actions run: 21990045205
Environment URL: review-environment | pipeline

[navalnica]

phrase "partially fixes #" will likely close original issue. maybe rephrase?

[bahdan111]

/deploy-review

GitHub actions run: 22065546129
Environment URL: review-environment | pipeline

[bahdan111]

/deploy-review

GitHub actions run: 22066587191
Environment URL: review-environment | pipeline

[navalnica]

there is a bug in statgpt.common.common.models.Channel:

• it specifies embedding model to be used by a channel, in llm_model field (field name is incorrect. and it likely led to the error below)
• the default value for llm_model field uses LLM model, not embedding model:

llm_model: Mapped[str] = mapped_column(default=langchain_settings.default_model.value)

I suggest we use:

llm_model: Mapped[str] = mapped_column(default=langchain_settings.embedding_default_model.value)

and delete default value in EmbeddingsModelConfig:

class EmbeddingsModelConfig(BaseModelConfig):
    """Config for embeddings models."""

    deployment: EmbeddingModelsEnum = Field(
        description="The deployment of the model in DIAL"
    )

@Fedir-Yatsenko what do you think?

[bahdan111]

/deploy-review

GitHub actions run: 22094853247
Environment URL: review-environment | pipeline

[bahdan111]

there is a bug in statgpt.common.common.models.Channel:

• it specifies embedding model to be used by a channel, in llm_model field (field name is incorrect. and it likely led to the error below)
• the default value for llm_model field uses LLM model, not embedding model:

llm_model: Mapped[str] = mapped_column(default=langchain_settings.default_model.value)

I suggest we use:

llm_model: Mapped[str] = mapped_column(default=langchain_settings.embedding_default_model.value)

and delete default value in EmbeddingsModelConfig:

class EmbeddingsModelConfig(BaseModelConfig):
    """Config for embeddings models."""

    deployment: EmbeddingModelsEnum = Field(
        description="The deployment of the model in DIAL"
    )

@Fedir-Yatsenko what do you think?

shall we move to separate PR, I would also like to change misleading name "llm_model" and we would have to migrate db after. @navalnica

[Fedir-Yatsenko]

shall we move to separate PR, I would also like to change misleading name "llm_model" and we would have to migrate db after. @navalnica

Let's first create an issue and then discuss it. If you want to change this, let's remove this column. Essentially, this parameter should be part of the Data Query config, not the channel.

[navalnica]

there is a bug in statgpt.common.common.models.Channel:

• it specifies embedding model to be used by a channel, in llm_model field (field name is incorrect. and it likely led to the error below)
• the default value for llm_model field uses LLM model, not embedding model:

llm_model: Mapped[str] = mapped_column(default=langchain_settings.default_model.value)

I suggest we use:

llm_model: Mapped[str] = mapped_column(default=langchain_settings.embedding_default_model.value)

and delete default value in EmbeddingsModelConfig:

class EmbeddingsModelConfig(BaseModelConfig):
    """Config for embeddings models."""

    deployment: EmbeddingModelsEnum = Field(
        description="The deployment of the model in DIAL"
    )

@Fedir-Yatsenko what do you think?

shall we move to separate PR, I would also like to change misleading name "llm_model" and we would have to migrate db after. @navalnica

let's change the varname in another PR

[bahdan111]

/deploy-review

GitHub actions run: 22131897847
Environment URL: review-environment | pipeline

[navalnica]

created follow-up issue to refactor langchain configs #150