How to Use Claude (Opus & Sonnet) Inside Sandbox AI

Claude is the model we reach for when the work is long, nuanced or legally sensitive — 200,000-token context, unusually careful writing, and agentic behavior that actually follows multi-step instructions. In Sandbox it sits one click away from every other frontier model, so you use each tool where it's genuinely strongest.

Where Claude wins

Long-document work — contracts, research reports, books. Opus ingests and reasons across entire documents without losing the thread.
Nuanced writing — editing that preserves authorial voice, diplomatic business comms, sensitive topics. It over-hedges less than its reputation suggests and rewrites less than GPT-5.
Agentic coding — multi-file changes, test-driven loops, tool use. This is the model our Code Agent routes most implementation steps to.
Instruction-following under pressure — Claude is notably hard to jailbreak into ignoring formatting rules, which matters in production pipelines.

Claude modes inside Sandbox

Claude Opus — flagshiphighest quality

Claude Sonnet — balanced2× cheaper

Extended context — 200K tokenswhole-repo / whole-book

Artifacts — live rendered outputcode · HTML · SVG

Claude variants available on Pro and above

Artifacts render live inside the workspace: ask Claude for a landing page, a chart or a small tool and it appears in a side panel you can run, edit and export — the same experience as Claude.ai, without a separate subscription.

Pair Claude with other models

The comparison mode is where Claude gets interesting. Draft the same brief with Claude and GPT-5 side by side and you'll quickly develop a feel for which model to route each task type to — Claude for the careful rewrite, GPT-5 for the punchy headline list, Gemini when the input is a video. Teams that do this for two weeks stop paying for the models they don't need and keep the ones they do.

⌘Analyze this 40-page contract. List every clause that favors the vendor, with page references.

⌘Refactor this module to remove the circular dependency. Keep the public API identical. Show the diff.

⌘Edit my draft below. Change nothing about my voice — only fix logic gaps and cut 15%. [paste]

Tip: for very large files, upload once and reuse across models in the same workspace — Sandbox caches the document so you can ask Claude, then Gemini, then GPT-5 the same question without re-uploading.

Why not just use claude.ai directly?

Claude.ai gives you Claude, full stop. The moment your work needs a second opinion — a punchier headline from GPT-5, a video summarized by Gemini — you're back to juggling logins and separate invoices. Sandbox installs as one Windows app: Claude's 200K context sits next to GPT-5 and 19 other models, billed once, at a fraction of what Claude Pro plus ChatGPT Plus costs separately.

Using Claude Opus and Sonnet inside Sandbox

Where Claude wins

Claude modes inside Sandbox

Pair Claude with other models

Why not just use claude.ai directly?

Claude + 20 other models. One plan.

Keep reading