Claude AI 2026: Complete Guide to Anthropic Products, Features and Sub-Products

Claude AI 2026: Complete Guide to Anthropic Products, Features and Sub-Products

You've heard about Claude, you may already be using it — but you're not sure exactly which features are available on which plan, or how Projects and Artifacts actually change your day-to-day workflow. That's understandable: the Anthropic ecosystem has evolved quickly and the official documentation scatters information across many pages.

This guide brings together everything you need to understand the Claude offering in 2026: plans and models, key features, use cases for architecture professionals, and the real limitations that rarely get mentioned.

Summary

1. The Claude product family: plans and models
2. Models: Haiku, Sonnet, Opus — which one for what
3. Projects — persistent context per client or project
4. Artifacts — code, documents, visualizations in the side panel
5. The 200,000-token context window
6. Vision and multimodality
7. Extended Thinking (Opus)
8. The Claude API for technical teams
9. Claude vs competitors
10. FAQ

The Claude product family: plans and models

Claude is not a single product — it's a family of offerings covering very different needs depending on whether you're an individual, a small firm, or a larger practice.

The free plan is fine for testing. Once you integrate Claude into a professional workflow — document processing, file analysis, technical content generation — the Pro or Teams plan quickly becomes necessary.

Important limitation: even on Pro, usage quotas are not unlimited — they are simply much higher than on the free plan. During high-load periods on Anthropic's servers, Pro users have priority over free users (official claude.ai documentation).

Models: Haiku, Sonnet, Opus — which one for what

Anthropic offers three model tiers, each with a different speed/capability profile.

Claude Haiku is the fastest and cheapest. Suitable for simple, repetitive tasks: classifying documents, extracting structured data, answering short questions. On the API, it's the default choice for applications where speed matters more than reasoning depth.

Claude Sonnet is currently the default model on claude.ai. It offers a good balance between speed and capability: report writing, contract analysis, Dynamo script generation for Revit, long structured responses. For most professional uses, Sonnet is the right compromise.

Claude Opus is the most capable model. It is slower and uses more tokens, but handles tasks requiring multi-step reasoning, deep analysis or structured creativity significantly better. A project manager asking Claude to analyze a 300-page tender document, identify contradictions between lots and produce a structured risk report will get a noticeably better result with Opus than with Sonnet. Available on Pro and above.

What Anthropic doesn't say clearly: Opus performance on long tasks degrades toward the end of context, as with all current large language models. On very long documents, the first sections are analyzed more thoroughly than the last.

Projects — persistent context per client or project

Projects is arguably the most useful feature for architects and professionals managing several clients simultaneously.

A Project is a workspace in which you can:

• Upload documents (brief, specifications, regulations, contracts) that remain available across all conversations in that Project
• Define custom instructions (system prompt) specific to that context
• Retrieve the conversation history linked to that client or project

Concrete use case: an architect opens a Project "Pine Residence — Developer Nexity". They upload the 80-page brief, the developer's energy performance requirements and their own meeting notes. In every subsequent conversation within that Project, Claude already knows the context: it can answer "what are the habitable floor areas on the ground floor?" without the architect re-uploading anything.

Real limitation: Projects are per-user on the Pro plan. To share them with a colleague at the practice, you need the Teams plan. Also, documents uploaded to a Project are subject to the same context window limits as normal conversations.

To go further on Project organization strategies, see our article on prompt engineering as a key skill in 2026.

Artifacts — code, documents, visualizations in the side panel

When Claude generates code, an HTML table, a structured document or an SVG diagram, the output appears in a side panel called an Artifact. The practical advantage: you see the result directly, without having to copy and paste into another tool.

For HTML/CSS/JS code, the Artifact offers an interactive browser preview. For Python or Ruby scripts, it displays the code with syntax highlighting and allows one-click copying.

What this changes for architects: an architect who asks Claude to write a Dynamo script to automate element renaming in Revit sees the script appear in the Artifact, ready to be copied into the Dynamo environment. They can request modifications, see the corrected version in the same panel, and iterate quickly.

Limitations to know: the Artifact is not a full execution environment. HTML preview works well for static pages, but not for code requiring a server or external dependencies. For Python with NumPy or Pandas, you'll see the code but can't execute it in the interface — you need to export it to your own environment.

The 200,000-token context window

200,000 tokens correspond roughly to 150,000 words, approximately 400 pages of dense text. This is the maximum context window available on Claude (official Anthropic documentation, 2026).

In practice, this means you can load into a single conversation:

• A complete tender package for a public procurement
• A local planning code running several hundred pages
• A set of contracts, amendments and site notes for an ongoing project

What you need to keep in mind: a large context window does not guarantee that Claude will process all information with the same quality. Research on "lost in the middle" degradation (Stanford NLP, 2023) shows that language models tend to process information better at the beginning and end of a long context. If your most critical document sits in the middle position, results may be less precise. A good practice is to place critical information at the start of the context.

Vision and multimodality

Claude accepts images as input: photos, screenshots, PDFs converted to images, floor plans, sections, elevations.

What it does well: analyzing a site plan image and commenting on circulation, identifying relationships between spaces, suggesting program adjustments. If you photograph a hand-drawn sketch and upload it, Claude can produce a structured description or identify contradictions with the written brief.

What Claude cannot do: read IFC, DWG or RVT files directly. These binary formats require a prior export to image or text (IFC to JSON, for example). Claude also has no real-time internet access — everything it knows comes from its training data, with a knowledge cutoff date.

Extended Thinking (Opus) — for complex tasks

The Extended Thinking feature allows Claude Opus to "think out loud" before responding: it unfolds its reasoning step by step, visible in a reasoning block, before producing its final answer.

Useful for tasks where reasoning quality matters more than speed: architectural feasibility analysis, arbitrating between contradictory scenarios, checking the consistency of a brief across multiple aspects simultaneously.

Limitations: Extended Thinking consumes significantly more tokens than a direct response. On the API, the cost is higher. And as with all large models, reasoning quality depends on the quality of the information provided — garbage in, garbage out.

For an introduction to autonomous AI agents that use Extended Thinking in more complex reasoning loops, see our dedicated article.

The Claude API for technical teams

The Claude API (available at api.anthropic.com) allows you to integrate Claude directly into your internal tools, without going through the claude.ai interface.

Architecture practices with a technical team can use it to:

• Build an internal tool that automatically processes responses to consultations (written document analysis, per-lot requirement summaries)
• Create a specification writing assistant based on your own document templates
• Automate BIM export analysis (IFC to JSON) to identify naming inconsistencies

The API supports vision, function calling, streaming and context caching — the last feature reduces the cost of requests that repeat the same long context, for example a reference brief used across many different analyses.

What slows adoption in practices: the API requires development skills (Python, JavaScript). It is not a no-code tool. For practices that want to get started without internal technical resources, no-code automation training remains a useful intermediate step.

To understand how to fund this type of technical training, see our guide on OPCO/FIFPL funding for AI training.

Claude vs competitors — where it's strong, where it falls short

Where Claude outperforms its direct competitors:

• Complex, long instruction-following: Claude respects precise editorial or technical constraints across long texts where other models drift
• Long documents: 200,000-token context vs 128,000 for GPT-4o (OpenAI, product documentation)
• Reluctance to hallucinate regulatory citations — Claude tends to signal uncertainty rather than inventing normative references, which matters in architecture where a misquoted standard can be costly
• French-language writing quality: field feedback consistently shows Claude Opus produces noticeably more natural French in technical documents

Where others do better:

• ChatGPT (OpenAI): broader integrations, DALL-E for image generation, native web access, customizable GPTs for specific workflows
• Gemini (Google): native Google Workspace integration, 1 million-token context window on Gemini 1.5 Pro — an advantage if you work in the Google ecosystem
• Mistral: European option, self-hostable (strict GDPR), good performance-to-cost ratio on the API

For a full comparison of the 4 major AI assistants for architects, see our dedicated article.

Structural limitations of Claude to know:

• No real-time internet access (unless via third-party integrations)
• No native image generation (unlike ChatGPT with DALL-E)
• No direct reading of proprietary BIM formats
• Quotas, even on Pro, can create friction during intensive work sessions

For a ChatGPT vs Claude comparison focused on architectural writing, see our detailed analysis.

FAQ

Is Claude Pro worth $20/month for an independent architect?

If you use Claude to process project documents (briefs, specifications, contracts), analyze tender responses or generate automation scripts, the answer is probably yes. The free plan hits its limits too quickly in real work sessions. However, if you only use Claude occasionally for short questions, the free plan may suffice.

Can I use Claude to analyze a site plan or section drawing?

Yes, within image format constraints. You need to export or photograph the drawing, then upload it to the conversation. Claude can comment on spatial relationships, flag areas to verify, or cross-reference the plan against written requirements. It does not read DWG, RVT or IFC files directly. For regulatory compliance analysis (fire safety, accessibility, energy performance), Claude's observations should be verified by a qualified professional.

What is the difference between Claude.ai and the Claude API?

Claude.ai is the general-purpose web interface. The API allows you to integrate Claude into your own applications or automated workflows. On the API, you control the model, temperature, system prompt and token management. It's more flexible but requires development skills. For a practice without a technical team, claude.ai with Projects is the natural starting point.

Does Claude keep my documents confidential?

Anthropic states that Pro users' conversations are not used to train models by default (official Anthropic documentation, 2026). The Enterprise plan includes additional contractual guarantees. If you process sensitive documents (contracts, client data), review the terms applicable to your plan before uploading.

Can training on Claude be funded through OPCO or similar schemes?

Yes, provided the training is delivered by a Qualiopi-certified organization. Educasium offers training in AI assistant use — including Claude — fundable through FIFPL for self-employed architects and via OPCO for employees of architectural practices. Details of these funding mechanisms are explained in our guide on OPCO/FIFPL funding for AI training.

Mastering Claude in your professional practice

Claude is a powerful tool for architecture professionals — for processing heavy documents, technical writing, brief analysis or script generation. But like any tool, its value depends on how you use it.

Structured training saves several weeks of trial-and-error learning. Educasium offers Claude and AI assistant training programs, Qualiopi-certified, fundable through FIFPL for self-employed architects and OPCO for employing practices.

Discover the AI training program for architects — 100% fundable, Qualiopi-certified.