Supported NLP Models
There are multiple ways to access the NLP models a study will need.
For example, a hypothetical study__nlp_gpt_oss_120b task that uses the GPT-OSS 120b model, can use Azure, Bedrock, or a local on-prem version of the model.
Below, you can see a support matrix of the models and platforms that Cumulus ETL supports. Individual study support might be more limited, depending on the study design. But these are all the models that a study could make use of:
| Azure | Bedrock | Local | |
|---|---|---|---|
| Claude Sonnet 4.5 | ❌ | ✅ | ❌ |
| GPT 3.5 | ✅ | ❌ | ❌ |
| GPT 4 | ✅ | ❌ | ❌ |
| GPT 4o | ✅ | ❌ | ❌ |
| GPT 5 | ✅ | ❌ | ❌ |
| GPT OSS | ✅ | ✅ | ✅ |
| Llama4 Scout | ✅ | ✅ | ✅ |
Using Azure
- Pass
--provider=azurewhen running NLP (e.g.cumulus-etl nlp --provider=azure ...) - Set
AZURE_OPENAI_API_KEYandAZURE_OPENAI_ENDPOINTenvironment variables
Using Bedrock
- Pass
--provider=bedrockwhen running NLP (e.g.cumulus-etl nlp --provider=bedrock ...) - Set the appropriate AWS environment variables, like
AWS_PROFILEorAWS_ACCESS_KEY_IDetc.