Microsoft AI Updates: October 15, 2026

1. GitHub Copilot Adds HydraFusion Max and Reworks Multi-Model Orchestration

Microsoft. GitHub updated HydraFusion, the Copilot option that orchestrates several models to produce one response, and added HydraFusion Max for harder tasks at higher reasoning effort; both are in research preview. The model pool is Claude Opus 5.5, GPT-6 Astra, GPT-6.1 Sol, and GPT-6 Luna, and each stage now gets explicit reasoning effort (medium for drafts, high for review). The cascade workflow adds a judging step that requests a repair only when needed, and the critique workflow now runs parallel critic models that inspect the workspace with tools rather than relying on text summaries. It is available in VS Code, the Copilot app, and Copilot CLI on Pro, Pro+, Business, and Enterprise plans with preview features enabled; premium request consumption was not stated. Source

2. Azure Content Understanding Improves Field Confidence Scores and Publishes a Calibration Lab

Microsoft. Microsoft said the latest Azure Content Understanding release integrates grounding more deeply into extraction, cutting average inference token usage by up to 28%, and that a refreshed confidence model improved AUROC by up to 14%. Confidence scores from 0 to 1 come from a separate service-level model rather than the LLM’s self-reported certainty, and v1 still returns no confidence for classifications. Microsoft recommends calibrating per-field thresholds on held-out, human-checked documents and published a notebook using 1,000 receipts from CORD v2 that outputs a field-level review routing table. In one insurance deployment, a calibrated policy flagged 41% fewer values for review across 22,000 fields while still catching more than 80% of known extraction errors. Source

3. Content Understanding 2.0 Preview Benchmarks Higher Accuracy at Lower Cost on US Tax Forms

Microsoft. Microsoft published benchmark results for Azure Content Understanding 2.0 Preview prebuilt analyzers across 19 US tax-form categories, reporting a mean F1 of 99.32% versus 97.41% for 1.0 GA and 96.71% for Azure Document Intelligence. Modeled cost was $8 per 1,000 pages versus $16.23 for 1.0 GA and $10 for Document Intelligence, with latency 51% lower than 1.0 GA but about 10% higher than Document Intelligence. The post introduces Advanced Contextualization for prebuilt analyzers and a 1099 Combo analyzer in preview that combines layout signatures, domain rules, and LLM understanding to assign values to the correct row and field. Cost figures are estimates, and Microsoft advises using the API’s usage output with current pricing. Source

4. GitHub Push Protection Now Blocks Anthropic API Keys by Default

Microsoft. GitHub added secret scanning detectors for anthropic_service_api_key and anthropic_user_api_key and enabled push protection for both by default on repositories with secret scanning, including free public repositories. Service keys found in public repositories are reported to Anthropic for revocation, while user keys generate alerts in both public and private repositories. The update also adds Salesforce Marketing Cloud OAuth2 tokens and Cloudsmith API keys. As agent workflows put model API keys into more repositories and config files, default blocking closes a common leak path. Source