Daily AI brief · July 23, 2026
Claude Now Available Natively Inside Copilot; Open-Weight Models Surge
Anthropic's Claude Sonnet 5 is live in Microsoft 365 Copilot, Google shipped three new Gemini models, Kimi K3 open weights drop later this week, and new Microsoft agent certification paths signal agentic AI is moving from optional to core.
Claude Sonnet 5 Now Native in Microsoft 365 Copilot (Microsoft 365 Copilot Blog)
Claude Sonnet 5 is rolling out now in Microsoft 365 Copilot, available in Copilot Cowork and PowerPoint .
Anthropic describes Sonnet 5 as a more capable agentic Sonnet model, with performance close to that of Opus 4.8, but at lower prices . You now have genuine model choice inside your existing workflow, not just brand consistency.
What it means for SMBs / MSPs: Your clients with Microsoft 365 can switch between Microsoft's models and Claude on a per-task basis for complex analysis or autonomous document work—no platform switch required.
Introductory pricing runs through August 31, 2026, at $2 per million input tokens and $10 per million output tokens , making A/B testing affordable.
Google Ships Gemini 3.6 Flash and Flash-Lite (Build Fast with AI)
Google released three new Gemini models on July 21: Gemini 3.6 Flash, Gemini 3.5 Flash-Lite, and Gemini 3.5 Flash Cyber, a security-tuned variant. Flash is the tier where most production volume actually runs, since the majority of enterprise AI calls are routine tasks that do not need flagship reasoning .
What it means for SMBs / MSPs: The efficiency tier got crowded. If you're evaluating Gemini for customer support or routine workflows, the Flash family now directly competes on speed and cost with Claude Sonnet 5 and DeepSeek V4 Pro. Benchmark the three before your next client proposal.
Kimi K3 Open Weights Ship July 27 (AI Weekly, Moonshot AI)
Kimi K3 was released by Moonshot on July 16, 2026, with the complete model weights scheduled to be opened on July 27 .
With 2.8 trillion parameters, a 1M token context, always-on inference, native multimodal capabilities, and MoE architecture , this is the largest open-source LLM to date.
What it means for SMBs / MSPs:
K3 carries 5x to 17x the output price of DeepSeek V4 Pro, with no downloadable weights until July 27 . If you're building agents that need visual understanding and your clients have large document volumes, wait three days to test K3 locally. For tight cost constraints, stick with DeepSeek V4.
Microsoft Shifts Agent Certifications to Agentic AI (Microsoft Partner Center)
New certification requirements, including Agentic AI Business Solutions Architect (AB-100) and AI Agent Builder Associate (AB-620) , have replaced the older Copilot certifications.
Copilot Cowork is Microsoft's agentic AI teammate, built to carry out real, multi-step tasks rather than just answer questions .
What it means for SMBs / MSPs: If you certify in this space, Microsoft is signaling that single-prompt Copilot skills are yesterday's game. Agents are the growth track.
Customers can submit their agents built in Agent Builder to the Agent Store after admin review and approval, so they can be discovered and used by others
—your custom workflows become sellable.
EU AI Act Transparency Guidelines Published (European Commission)
The July 2026 action plan on Cybersecurity and AI sets out a coordinated approach to help Member States, businesses and public authorities address cybersecurity and resilience challenges posed by the most advanced AI models .
What it means for SMBs / MSPs: If your clients sell into Europe or hold European data, compliance documentation is no longer "nice to have." Prepare model cards and data lineage audits now—procurement teams will ask.
Test Claude Sonnet 5 Against Your Current Copilot Setup
Claude is now a selectable model inside Copilot Chat, alongside Microsoft's own models. Users can choose Claude for complex analysis, document understanding and structured content generation, then switch back for everyday tasks .
Who it's for: MSP tech leads and AI champions evaluating agent performance on client workflows.
How long a first test takes: 20 minutes. Enable Claude in your tenant admin panel, pick one complex document task (legal contract analysis, financial summary, code review), run it against Claude Sonnet 5 and the default model side by side, measure latency and cost.
One concrete first step: Create a test team in Microsoft 365, enable Claude as an option, and run your three most common Copilot queries. Screenshot costs and response times. Use that data when pitching model choice to clients as a productivity unlock.
Ask whoever runs your Microsoft 365 whether Claude is on in Copilot, and who picks the model. Claude Sonnet 5 is now native in Microsoft 365 Copilot, so your team may already have a second model without adding a vendor. Decide which tasks, such as contract review or redaction, are allowed on which model. Add that to your staff AI rules.
*For MSPs:* For legal, compliance and healthcare clients wary of Copilot, show them model choice inside the platform they already use before any wider rollout.
Don't chase Kimi K3's benchmarks without waiting for the weights on July 27.
Until then, K3 is usable only through the API and Kimi apps . The published benchmarks are Moonshot's own; independent verification isn't complete. Pay $15 per million tokens on the API for a model you can't self-host yet, or wait 72 hours to download it and run real cost models. The hype will cool once the weights land and the community stress-tests the numbers.
Open-source LLMs are now competitive vectors for enterprise architecture, not just cost plays.
Moonshot AI's Kimi K3, DeepSeek V4 Pro, and Zhipu AI's GLM-5.2 are all sparse Mixture-of-Experts models with one-million-token context windows aimed at coding and agent tasks .
China is narrowing the model-capability gap while operating under hardware constraints, and that has consequences for model pricing, inference deployment, hardware diversification, and the effectiveness of U.S. export controls . For SMBs and MSPs, this means you can now run a 3-trillion-parameter agent locally—not because you must, but because the value per dollar tilts that way for many workflows. Your infrastructure decisions are shifting: cloud-based APIs for burst, self-hosted agentic systems for predictable, repeating workloads. That flexibility is new, and it's worth testing with your clients now.