ScrapeGraphAI/Scrapegraph-ai · error · ValueError

Provider {llm_params["model_provider"]} is not supported.

Error message

Provider {llm_params["model_provider"]} is not supported.
                If possible, try to use a model instance instead.

What it means

Raised when _create_llm cannot determine the provider: the 'model' string in llm_params does not appear in any provider's entry of models_tokens, so the list of possible providers is empty and the fallback lookup fails. The message prints the (unset or wrong) model_provider value, which is confusing but means 'unknown model name'.

Source

Thrown at scrapegraphai/graphs/abstract_graph.py:191

            "fireworks",
            "clod",
            "togetherai",
            "xai",
            "minimax",
        }

        if "/" in llm_params["model"]:
            split_model_provider = llm_params["model"].split("/", 1)
            llm_params["model_provider"] = split_model_provider[0]
            llm_params["model"] = split_model_provider[1]
        else:
            possible_providers = [
                provider
                for provider, models_d in models_tokens.items()
                if llm_params["model"] in models_d
            ]
            if len(possible_providers) <= 0:
                raise ValueError(
                    f"""Provider {llm_params["model_provider"]} is not supported.
                                If possible, try to use a model instance instead."""
                )
            llm_params["model_provider"] = possible_providers[0]
            logger.info(
                "Found providers %s for model %s, using %s. "
                "If it was not intended please specify the model provider in the graph configuration",
                possible_providers,
                llm_params["model"],
                llm_params["model_provider"],
            )

        if llm_params["model_provider"] not in known_providers:
            raise ValueError(
                f"""Provider {llm_params["model_provider"]} is not supported.
                             If possible, try to use a model instance instead."""
            )

View on GitHub (pinned to 532dfffbf6)

Solutions

Check the exact model string keys in scrapegraphai/models/models_tokens.py and use one of them verbatim.
Upgrade scrapegraphai (uv sync / pip install -U scrapegraphai) so newly added models are recognized.
For custom models, pass a pre-built 'model_instance' plus 'model_tokens' to bypass name-based lookup.
Set 'model_provider' explicitly to a supported provider if the model name genuinely exists under it.

Example fix

# before
config = {'llm': {'model': 'gpt4-o', 'api_key': key}}

# after
config = {'llm': {'model': 'gpt-4o', 'model_provider': 'openai', 'api_key': key}}

Defensive patterns

Strategy: validation

Validate before calling

from scrapegraphai.models.models_tokens import models_tokens
model = config['llm']['model']
known = {m for prov in models_tokens.values() for m in prov}
assert model in known, f"model '{model}' not in models_tokens; fix the name or pass model_instance"

Type guard

def is_known_model(model: str) -> bool:
    from scrapegraphai.models.models_tokens import models_tokens
    return any(model in prov for prov in models_tokens.values())

Try / catch

try:
    graph = SmartScraperGraph(prompt=p, config=config)
except ValueError as e:
    if 'is not supported' in str(e):
        config['llm']['model_instance'] = build_client(config['llm'])
        config['llm']['model_tokens'] = 32000
        graph = SmartScraperGraph(prompt=p, config=config)
    else:
        raise

Prevention

Copy model names verbatim from models_tokens.py.
Pin and periodically upgrade scrapegraphai so new model names are recognized.
For custom endpoints, standardize on model_instance + model_tokens.

When it happens

Trigger: Setting config llm {'model': 'some-unknown-or-typo'd-model'} without a valid model_provider, or with model_provider omitted so the code tries to infer the provider from models_tokens and finds no match.

Common situations: Typos in model names ('gpt-4o-mini ' with whitespace, 'claude-3.5-sonnet' vs the exact key); brand-new models released after the installed scrapegraphai version; custom/self-hosted model names that are not in the built-in token tables.

Related errors

AI-assisted analysis of ScrapeGraphAI/Scrapegraph-ai@532dfffbf6 (2026-08-28). Data as JSON: /api/errors/47d5b0c62d01949f. Report an issue: GitHub.