Microsoft adds GPT-5.6 Terra and advanced NL2SQL to Fabric data agents
Microsoft will roll out Advanced NL2SQL, advanced DAX generation, and the GPT-5.6 Terra model to the standard runtime of Fabric data agents between November 2 and November 10, 2026. The changes are automatic; users cannot enable them manually. Microsoft says internal tests show NL2SQL accuracy rising from 78.8% to 90.9% and P95 processing time dropping from 46.8 seconds to 32.9 seconds, though…
Key points
- Microsoft deploys Advanced NL2SQL, advanced DAX generation, and GPT-5.6 Terra to Fabric data agents from November 2 to November 10, 2026
- Internal tests show NL2SQL accuracy rising from 78.8% to 90.9% and P95 latency dropping from 46.8 seconds to 32.9 seconds
- Users should baseline and re‑test 10 to 20 daily questions, checking queries, clarification prompts, and processing times
Organizations are advised to save baseline responses, then re‑test 10 to 20 daily questions during the rollout, checking generated SQL or DAX, clarification prompts, and latency. The guidance stresses comparing query structure and referenced tables, not just answer text, especially for metrics like sales or inventory. Teams, Microsoft 365 Copilot, or Copilot Studio users should run real‑world expressions and note any discrepancies before expanding automation.
The update aims to improve accuracy but does not guarantee compatibility with existing settings, and Microsoft’s internal figures do not assure the same outcomes for every customer. Companies should record their own accuracy, processing time, and human verification effort before deciding to adopt the new runtime fully.
Microsoft Fabric AI Data Analysis Updates Automatically. Why You Should Re-test by November 10th
note.com · 2 November 2026
Loading the full article…
This text was published by note.com and written by noahs_nalog/AIツール実務レター/あなたのモチベーション. It is reproduced here with attribution so you can read it in full; the rights remain with the publisher. Read it at the source ↗
The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.
More in Agents & Tools
All →- Descript runs live MCP builds with Claude, ChatGPT and Codex · 1 src
- AI web search works best for common queries · 1 src
- OpenAI, Meta, Anthropic launch AI assistants that act before users ask · 1 src
- Meta releases muse agent, Amazon blocks it · 1 src
- DeepSeek launches Super-Agent to automate developer tasks without human input · 6 src
Comments
via GitHub Discussions