Context & Output Capacity: it has 1 million token context window — the first Opus model to support this (in beta via the developer platform); and 128K token output — a significant bump in how much content can be generated in a single response.

Agent Teams: a new capability where teams of agents can split larger tasks into segmented jobs and work together, in opus4.5 it’s parent-child, agent-subagent structure, while opus4.6 we’ll have real team of agents that are parallel.

Zero-day vulnerability detection: Anthropic reports the model found over 500 previously unknown zero-day vulnerabilities in open-source code out-of-the-box, each validated by Anthropic staff or external security researchers.

Finance: Tops the Finance Agent benchmark. Can combine regulatory filings, market reports, and internal data for end-to-end financial analysis.

Powerpoint integration: Research preview of direct integration into Microsoft PowerPoint — can read existing layouts, fonts, and templates, then generate or edit slides while preserving design elements.

Enterprise focus: Built around three outcomes: finding information, analyzing it, and producing finished outputs (documents, spreadsheets, presentations) closer to production-ready on the first attempt.

So overall, it seems Anthropic is more ambitious, expanding from being the best coding assistant to aiding real financial enterprise knowledge workers in their job functionalities now.

To test it out, I just asked claude to do a in-depth analysis of Tesla and generate user-friendly documents in pdf and ppt forms:

In about 3 minutes, it wrote the full-blown TSLA-analysis codes and generated quite professional pdf and ppt:

I will upload a short youtube video to showcase the full process.