Models Used by Solid
A current inventory of the AI models Solid uses across query generation, semantic modeling, search, and background enrichment.
Current model inventory — September 16, 2026
Solid uses different AI models across query generation, semantic modeling, search, and background enrichment. The tables below summarize which models support each product area.
Query generation (Text2SQL)
Solid uses gpt-5.5 by default to translate natural-language questions into SQL. Additional models support alternative query generation, correction, optimization, and candidate selection.
| Task | Models used |
|---|---|
| Initial SQL generation | gpt-5.5 (default) |
| Optional ensemble generation—producing alternative SQL candidates | o3-mini, gpt-chat-latest (5.3 chat), gpt-5.4, gpt-5.5 |
| Correct queries that time out or return empty results | o3-mini |
| Correct other query failures or SQL syntax | gpt-4.1 |
| Expand or rewrite SQL | gpt-5.5 |
Optimize SQL using execution plans (EXPLAIN) | gpt-4.1 |
| Select among candidates and perform lightweight SQL tasks | gpt-5-mini, o3-mini |
Models across the platform
| Product area | How models are used |
|---|---|
| Clarification and routing | gpt-4.1 supports clarifying questions, glossary tasks, and semantic model selection. gpt-4.1-mini handles asset classification and query variants. o3-mini supports more complex filtering and clarification. |
| Semantic model creation, editing, and quality checks | gpt-4.1 and gpt-5-mini support model generation and save workflows. gpt-4.1-mini provides assistive completions and lightweight editing. o3-mini, gpt-chat-latest (5.3 chat), and gpt-4.1-mini support questions about semantic models and verification. |
| Search and retrieval | text-embedding-3-small creates embeddings for retrieval. Cohere-rerank-v4.0-fast reranks retrieved results to help determine which context is provided to the language model. |
| Background metadata enrichment | gpt-5 is the default for offline processing, including batch workloads. gpt-5-mini and gpt-4.1-mini handle lighter enrichment steps. These processes help shape the metadata customers later see in Solid. |
| BI export and enrichment | Primarily o3-mini and gpt-4.1 support export and enrichment workflows for BI tools such as Looker. |
How to read this inventory
Model usage depends on the workflow and processing step; a request may use only a subset of the models listed. Optional ensemble variants apply when that generation path is used. Here, semantic models refers to Solid's definitions of data and business logic, while AI models refers to the models performing the tasks above.
Updated 12 days ago