AI Chatbot Platform
Layer-by-layer comparison of ChatGPT, Claude, Gemini, and DeepSeek — from model internals to client UI and infrastructure.
This is a living document. Update it when major model versions, protocols, or platform architecture change.
Platform Stack
| Dimension |  ChatGPT |
 Claude |
 DeepSeek |
 Gemini |
|---|---|---|---|---|
| Company | OpenAI | Anthropic | DeepSeek AI | Google DeepMind |
| Type / License | Proprietary — closed weights, closed API | Open weights (MIT / Apache 2.0 per model); closed API available | Proprietary — closed weights, closed API | |
| Model family (Jun 2026) | GPT family; Instant / Thinking / Pro auto-routing | Opus / Sonnet / Haiku family | DeepSeek family; open-model-first | Gemini family (Flash + Pro) |
| Architecture | Decoder-only Transformer; internals undisclosed | Decoder-only Transformer; open weights | Decoder-only Transformer; confirmed MoE top-k routing | |
| Multimodality | Text, image, audio, voice | Text + vision | Text; focus on reasoning and coding | Native interleaved: text, image, audio, video |
| Context window | ~400K tokens | 200K+ (tier-dependent) | 1M tokens | |
| Output + transport | Markdown + SSE / JSON deltas — de facto standard (Gemini additionally uses gRPC / protobuf internally) | |||
| Render path | Markdown → AST → React components — no major divergence | |||
| Rich UI / agents | Apps SDK + Canvas over MCP | Artifacts + Cowork / Design; MCP origin | — | Canvas + Workspace depth; A2UI + MCP |
| Tool standard | MCP — variance is ecosystem maturity and packaging, not protocol direction | |||
| Client | Web React / TypeScript + native apps — depth varies by ecosystem | |||
| Infra | Azure | AWS Trainium + Google TPU | — | Google TPUs |
Capabilities
Beyond chat, each platform exposes interactive tools and rich rendering. What the app can do in a conversation — search the web, run code, render diagrams, preview UI — increasingly matters as much as raw model quality. Snapshot below; these features ship and change fast.
| Capability | ChatGPT |
Claude |
DeepSeek |
Gemini |
|---|---|---|---|---|
| Web search | ✅ with citations | ✅ | ✅ | ✅ Search grounding |
| Code execution | ✅ Python sandbox | ✅ analysis tool | ⚠️ code-gen only | ✅ Python |
| Live editor | Canvas | Artifacts | — | Canvas |
| Mermaid diagrams | ✅ in Canvas | ✅ in Artifacts | code only | ✅ in Canvas |
| SVG / HTML / React preview | ✅ Canvas | ✅ Artifacts | — | ✅ Canvas |
| Interactive maps | ✅ Apps (e.g. Mapbox) | ⚠️ via MCP | — | ✅ Google Maps |
| Image generation | ✅ | — | — | ✅ |
| Voice (in / out) | ✅ Advanced Voice | ⚠️ mobile | — | ✅ Gemini Live |
| File / data analysis | ✅ | ✅ | ⚠️ basic | ✅ |
| Apps / extensibility | ✅ Apps SDK + MCP | ✅ MCP connectors | — | ✅ A2UI + MCP |
Note: Rich-UI output renders through the live-editor surfaces — ChatGPT Canvas, Claude Artifacts, Gemini Canvas — which preview HTML/React, SVG, and Mermaid inline. Maps and other third-party widgets arrive through the Apps/MCP layer, not the base model.
AI Coding Clients
Claude Code, Codex, GitHub Copilot, and OpenCode solve a similar problem (AI-assisted coding), but they are client-layer products — not foundation model platforms.
| Dimension |  Claude Code |
 Codex |
 GitHub Copilot |
 OpenCode |
|---|---|---|---|---|
| Company | Anthropic | OpenAI | GitHub (Microsoft) | sst (community) |
| Type / License | Proprietary client — closed source; subscription required | Open-source client — MIT; subscription required for the API | Open-source client — MIT; BYOK: Ollama, any API provider, or GitHub-hosted models | |
| Model strategy | Tightly coupled to native model ecosystem | Multi-provider: GitHub-hosted models, BYOK, or local (Ollama) | ||
| Interface | Agentic coding sessions in terminal / editor | VS Code / IDE chat + inline completion + agent flows | Terminal-first workflow | |
| Tool fit | Integrated coding-tool loops | MCP ecosystem alignment | MCP / open-tooling flexibility | |
| Layer | Application-layer clients above model platforms | |||
Key Takeaways
- The middle layers have converged: Markdown output, SSE + JSON delta streaming, Markdown → AST → React rendering, and MCP as the tool-calling standard.
- Real differences sit in model behavior, reasoning quality, context window reliability, product UX, and infrastructure strategy.
- DeepSeek is the only platform here with open weights — a meaningful distinction for self-hosting and reproducibility.
- Claude Code, Codex, GitHub Copilot, and OpenCode are coding clients built on top of these platforms, not independent model stacks.
- Most internal architecture details remain proprietary; treat vendor-unconfirmed claims as estimates.
Maintenance Note
Update this article when model families, default routing behavior, context windows, or protocol layers change.