xAI released Collections API on December 22, letting developers offload the messy work of document indexing and retrieval. Upload PDFs, Excel files, or codebases, and the service handles storage, parsing, and semantic search.
The pitch is straightforward: build RAG applications without managing vector databases. xAI uses OCR and layout-aware parsing to preserve document structure, so tables in spreadsheets and syntax in code files stay intact for the model. When files change, the system reindexes automatically.
Retrieval options include semantic search, keyword matching, or a hybrid mode that combines both with reranking. xAI's own benchmarks show Collections outperforming Gemini 3 Pro and GPT-5.1 on data extraction tasks in finance and legal domains, though the comparison comes with caveats: Gemini does not expose the actual retrieved files, so the metric measures files cited by Gemini rather than raw retrieved files. Independent verification hasn't appeared yet.
Pricing runs $2.50 per 1,000 searches, with file indexing and storage free for the first week. xAI says uploaded files aren't used for model training without consent.
The Bottom Line: Collections API removes RAG infrastructure overhead for developers willing to pay $2.50 per thousand queries and trust xAI's benchmarks.
QUICK FACTS
- Launch date: December 22, 2025
- Retrieval pricing: $2.50 per 1,000 searches
- File types: PDFs, Excel sheets, codebases
- Retrieval modes: semantic, keyword, hybrid
- Free trial: indexing and storage for first week
- Benchmark claims: company-reported, not independently verified
