Got an error when building a knowledge graph index with GPT-5

[breakaice]

Hi,

I tried to build a knowledge graph index using GraphRAG v2.6.0 with model: gpt-5 in my settings.yaml.
However, the process fails with the following error in the indexing-engine.log

Error code: 400 - {
'error': {
'message': "Invalid type for 'max_tokens': expected an unsupported value, but got null instead.",
'type': 'invalid_request_error',
'param': 'max_tokens',
'code': 'invalid_type'
}
}

GraphRAG version: 2.6.0
openai-python: 1.101.0
fnllm: 0.4.1

It works fine if I use model: gpt-4o, but always fails with model: gpt-5.

My .env has a valid key, and other configs (embedding: text-embedding-3-small) are fine.

[zedhaque]

I ran into similar errors initially; I’m now seeing a different one. To unblock your current issue, try the following:
• Use LiteLLM for the GPT-5 series.
• Update your settings.yml to include the required keys (e.g., provider, model, max_completion_token, etc.). Reference: https://microsoft.github.io/graphrag/config/yaml/

This is the error I got now, maybe I also need to update entity extraction prompt.

graphrag.logger.progress - extract graph progr--More--(70%)
2025-09-24 01:51:26.0326 - INFO - graphrag.logger.progress - extract graph progress: 15/16
2025-09-24 01:51:27.0083 - INFO - graphrag.logger.progress - extract graph progress: 16/16
2025-09-24 01:51:27.0087 - ERROR - graphrag.index.run.run_pipeline - error running workflow extract_graph
Traceback (most recent call last):
  File "/app/indexer/.venv/lib/python3.12/site-packages/graphrag/index/run/run_pipeline.py", line 121, in _run_pipeline
    result = await workflow_function(config, context)
             ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/app/indexer/.venv/lib/python3.12/site-packages/graphrag/index/workflows/extract_graph.py", line 50, in run_workflow
    entities, relationships, raw_entities, raw_relationships = await extract_graph(
                                                               ^^^^^^^^^^^^^^^^^^^^
  File "/app/indexer/.venv/lib/python3.12/site-packages/graphrag/index/workflows/extract_graph.py", line 95, in extract_graph
    extracted_entities, extracted_relationships = await extractor(
                                                  ^^^^^^^^^^^^^^^^
s/extract_graph/extract_graph.py", line 79, in extract_graphhrag/index/operation--More--(86%)
    entities = _merge_entities(entity_dfs)
               ^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/app/indexer/.venv/lib/python3.12/site-packages/graphrag/index/operations/extract_graph/extract_graph.py", line 103, in _merge_entities
    all_entities.groupby(["title", "type"], sort=False)
  File "/app/indexer/.venv/lib/python3.12/site-packages/pandas/core/frame.py", line 9190, in groupby
    return DataFrameGroupBy(
           ^^^^^^^^^^^^^^^^^
  File "/app/indexer/.venv/lib/python3.12/site-packages/pandas/core/groupby/groupby.py", line 1330, in __init__
    grouper, exclusions, obj = get_grouper(
                               ^^^^^^^^^^^^
  File "/app/indexer/.venv/lib/python3.12/site-packages/pandas/core/groupby/grouper.py", line 1043, in get_grouper
    raise KeyError(gpr)
KeyError: 'title'
2025-09-24 01:51:27.0097 - ERROR - graphrag.api.index - Workflow extract_graph completed with errors

[natoverse]

This should be resolved in 2.6.0; it was an issue with a changed response type from OpenAI that we needed to update to