Claude vs Gemini: reasoning depth vs the million-token bundle

Claude's case is quality of thought: the models people reach for when the writing or reasoning has to be right, with the most controllable memory in the business. Gemini's case is capacity and value: a million-token window on paid plans, a $4.99 floor, and the Google bundle. Both are linear chat apps with one vendor's models and unpublished usage limits. We build NodePad; the third column is for the people who want Claude's judgment on one message and Gemini's long-context stamina on the next, inside one workspace where the context is assembled by hand and visible. One-vendor loyalists should just buy their vendor. Everyone else is why this page has three columns.

dimension Claude Gemini NodePad
Price of entry Free; Pro $20/mo ($17 annual); Max $100 to $200/mo. Free; AI Plus $4.99/mo; AI Pro ~$20/mo; AI Ultra $100 to $200/mo; the lowest serious floor. Free on community models; paid from $5/mo.
Models Haiku, Sonnet 5, Opus 4.8, Fable 5; free tier gets Haiku and Sonnet. Fast, Thinking, and Pro labels; exact versions swap underneath them. Claude, GPT, and Gemini, picked per message, with exact model names.
Context window 200K on all consumer plans; approaching the limit triggers silent auto-summarization. 32K free, 128K Plus, 1M on Pro and Ultra, 5x Claude's consumer ceiling. The model's full window, spent on context you chose rather than the whole scrollback.
Who assembles the context Memory synthesis plus chat-search retrieval; you edit the summary, not the per-message context. Personalization from chat history and, opted in, Gmail and Drive. You. The path you built plus what you referenced or merged. Nothing hidden, nothing injected.
Branching Edit and regenerate keep hidden alternate versions behind small arrows; no overview of the paths. None; strictly linear. Fork any point onto a visible parallel path on the canvas.
Organizing real work Projects with knowledge bases, instructions, and per-project memory (auto-RAG beyond the window). No working folders shipped as of mid-2026; a sidebar list with pin and rename. Workspaces where the chats themselves sit on one canvas and can be merged.
Memory Best-in-class controls on every plan: view, edit, pause, reset, incognito. A personalization toggle; deletable activity items; no editable summary. None by design; what the model sees is on the canvas, not in a profile.
Beyond chat Artifacts with publishing, Claude Code bundled at $20, voice, MCP connectors. Gemini Live free, image and video generation, Workspace integration, storage and YouTube bundles. The canvas is the product; plugins add tools. Both columns beat us on breadth here.
Usage limits A conversation budget with no published numbers; chat and Claude Code drain one pool. Published as multipliers only (2x, 4x, 5x, 20x); absolute numbers nowhere. A visible meter; add usage or wait for the monthly reset.
Try NodePad free

No credit card to start. Every paid plan reaches every model.