Claude Sonnet 4.6 in Microsoft Foundry: 1M Context

Introduction: why this matters

Claude Sonnet 4.6 is now available in Microsoft Foundry, giving IT and engineering teams another “frontier-class” model option that targets large-scale enterprise workloads—without always needing the premium cost profile of top-tier models. For organizations building developer copilots, knowledge-work assistants, and automation agents, Sonnet 4.6 aims to balance intelligence, throughput, and governance-friendly deployment in a managed environment.

What’s new in Claude Sonnet 4.6 (in Foundry)

Massive context + large outputs

• 1 million token context window (beta), matching the extended context tier called out for Claude Opus 4.6.
• Up to 128K maximum output, enabling longer generated artifacts (reports, code changes, multi-step plans) in a single response.
• Practical upside: less “context chopping” for large codebases, long financial models, multi-document review, and long-running multi-turn agent workflows.

Adaptive thinking and “effort” controls

• Sonnet 4.6 introduces adaptive thinking, allowing the model to apply deeper reasoning only when needed.
• Effort parameters provide a way to tune the quality–latency–cost tradeoff for different tasks (e.g., quick triage vs. high-assurance analysis).

Developer-focused upgrade (vs. Sonnet 4.5)

Microsoft highlights Sonnet 4.6 as a direct upgrade to Sonnet 4.5 with minimal prompting changes required for most workflows. Improvements emphasized include:

• Stronger reasoning across broader code context
• Better understanding of complex codebases
• More reliable performance across iterative development cycles (feature build, refactor, debug, refine)

Improved knowledge work at scale

For enterprise content and analysis workflows, Sonnet 4.6 targets fewer revision cycles for:

• Report drafting and refinement
• Summarizing large document sets
• Structured business documentation
• Presentation and narrative generation

“Computer use” for browser automation

Sonnet 4.6 is described as Anthropic’s most capable computer use model yet, scoring 72.5% on OSWorld Verified. Key enterprise implications:

• Automates browser-based tasks without depending on an API (useful for legacy and SaaS tools with limited integration).
• Improved UI interaction precision (e.g., difficult click targets).
• Can move across web apps and complete multi-step tasks (e.g., read context, respond to a message, create a calendar event).

Impact for IT admins and platform teams

• Architecture choices: Sonnet 4.6 can serve as a cost-efficient default model for high-volume chat, document workflows, or as a sub-agent in multi-model pipelines.
• Automation reach: Browser-based automation expands what’s possible where APIs don’t exist, but increases the need for guardrails, auditing, and least-privilege access.
• Operational governance: Deploying via Microsoft Foundry positions teams to align model usage with enterprise expectations for governance, compliance, and operational tooling.

Action items / next steps

1. Pilot Sonnet 4.6 in Foundry for one high-context scenario (e.g., repo-wide code assistant, policy/document analysis) to validate the 1M context benefits.
2. Define “effort” profiles (low/medium/high) mapped to task classes to control cost and latency.
3. Evaluate computer-use automations in a sandbox: start with non-destructive workflows (read-only validation, QA UI checks) before enabling transactional actions.
4. Attend Model Mondays (Feb 23) for guidance from Anthropic on Opus 4.6 and Sonnet 4.6 architectures and enterprise deployment patterns.