# ChatRound — Full Documentation > Ask once. Compare answers from multiple AI models and quickly understand their differences and key takeaways. ## Overview ChatRound is a web application for asking one question to several AI models at once. You select the models and ask once. Each answers, then an automatic summary highlights key takeaways, differences, and gaps. You can choose targeted follow-ups when useful. It is hosted at https://chatround.ai. You can start on a free plan with a monthly allowance, subscribe to Pro for a larger one, or supply your own OpenRouter key or custom endpoint — that last way is unmetered and the provider bills you directly, with no ChatRound markup. ## Core Concepts ### Conversations A conversation is the unit of work: one thread holding your question, the models' answers, and any follow-ups. It carries its own set of selected models, its mode, and its web-search setting. Conversations are top level — there is no folder or workspace layer above them. ### Answers and comparison New conversations start with independent answers: models see the user's messages and their own answer history. Each answer wave is followed by a summary. Original answers and the summary are available in the same result tabs; a completed summary does not automatically switch the answer being read. Existing conversations retain their saved context-sharing setting. ### Optional follow-ups The summary may suggest targeted questions. Expand the optional follow-ups, choose recipients, edit the questions, or ask your own. Sending these questions lets the selected models respond with shared answer context, followed by an updated summary. Expanding the suggestions does not call models. Each additional model answer uses a turn. A single question and its summary can be a complete result. ### Moderator The moderator is a model too. It makes one structured call per round producing both the report you read and a brief for each participant, every brief ending in a specific question that model must answer next. GPT-5.6 Luna is the platform-funded default on both plans, selected for citation accuracy in ChatRound's production-format tests. When you use your own route, you can pin the moderator provider; otherwise ChatRound prefers a reachable moderator whose provider is not itself in the discussion. ## Models 29 models across eight sources are offered for new conversations. Superseded models remain available to existing conversations, as listed below. ### OpenAI - GPT-6 Astra — frontier, 1M context, web search, images, PDF, reasoning effort - GPT-6.1 Sol — flagship, 1M context, web search, images, PDF, reasoning effort - GPT-5.6 Terra — general, 1M context, web search, images, PDF, reasoning effort - GPT-6 Luna — fast, 1M context, web search, images, PDF, reasoning effort - GPT-5.4 Nano — lite, 400K context, web search, images, PDF, reasoning effort ### Anthropic - Claude Fable 5.1 — frontier, 1M context, web search, images, PDF, reasoning effort - Claude Opus 5.5 — flagship, 1M context, web search, images, PDF, reasoning effort - Claude Sonnet 5.5 — general, 1M context, web search, images, PDF, reasoning effort - Claude Haiku 4.5 — fast, 200K context, web search, images, PDF ### Google - Gemini 3.1 Pro — flagship, 1M context, web search, images, PDF, reasoning budget - Gemini 3.8 Flash — general, 1M context, web search, images, PDF, reasoning effort - Gemini 3.5 Flash-Lite — fast, 1M context, web search, images, PDF - Gemini 2.5 Flash-Lite — lite, 1M context, web search, images, PDF ### xAI - Grok 4.7 — flagship, 500K context, web search, images, reasoning effort - Grok 4.3 — general, 1M context, web search, images, reasoning effort - Grok 4.20 — long-ctx, 2M context, web search, images ### DeepSeek - DeepSeek V4 Pro 0813 — flagship, 1M context, reasoning effort - DeepSeek V4.1 Flash — fast, 1M context, reasoning effort ### GLM - GLM-5.3 — flagship, 1M context, reasoning effort (always on) - GLM-5V Turbo — vision, 200K context, images, reasoning toggle - GLM-4.7 — general, 200K context, reasoning toggle - GLM-5.3 Flash — fast, 1M context, images, always-on thinking; no off switch ### Kimi - Kimi K3 — flagship, 1M context, images, reasoning toggle - Kimi K2.6 — general, 256K context, images, reasoning toggle - Kimi K2 Thinking — fast, 256K context, always-on thinking; no off switch ### Qwen - Qwen3.8 Max 0902 — flagship, 1M context, images, reasoning effort (always on) - Qwen3.7 Plus — general, 1M context, images, reasoning toggle - Qwen3.8 Flash — fast, 1M context, images, reasoning toggle - Qwen3.5 9B — lite, 262.144K context, images ### Legacy conversation compatibility These models are retained for old conversations, not offered in the new-conversation picker. Their original ids, attachment capabilities and routes are preserved. - Claude Fable 5 → Claude Fable 5.1 - Gemini 3.6 Flash → Gemini 3.8 Flash - Claude Opus 5 → Claude Opus 5.5 - Claude Sonnet 5 → Claude Sonnet 5.5 - GPT-5.6 Sol → GPT-6.1 Sol - GPT-5.6 Luna → GPT-6 Luna - Grok 4.6 → Grok 4.7 - Grok 4.5 → Grok 4.7 - DeepSeek V4 Pro → DeepSeek V4 Pro 0813 - DeepSeek V4 Flash 0731 → DeepSeek V4.1 Flash - DeepSeek V4 Flash → DeepSeek V4.1 Flash - GLM-5.2 → GLM-5.3 - Qwen3.7 Max → Qwen3.8 Max 0902 DeepSeek 0813 uses a versioned OpenRouter id; native DeepSeek-compatible endpoints use the official deepseek-v4-pro and deepseek-flash (V4.1 Flash) API aliases. Prices in the product catalogue are baseline channel estimates, excluding caching, long-context bands and provider-specific overrides. Capabilities are per model, not per company, and describe what ChatRound's route to that model can deliver. The model picker only offers a capability on models that have it. PDF input is listed only where ChatRound's route can carry a document part, which is why the open-weight sources — reached over OpenAI-compatible chat completions — show images but not PDF even where the vendor's own API reads documents. ## Attachments Images (PNG, JPEG, WebP, GIF), PDFs, and plain-text files up to 10 MB each. An attachment is only sent to models that can read that kind of input. ## Web Search Web search is a per-model switch, available on the models whose provider exposes a search tool: the OpenAI, Anthropic, Google, and xAI models above. ## Languages The interface is available in English, Simplified Chinese, Japanese, Korean, Spanish, and Brazilian Portuguese. ## Plans and API Keys ChatRound can fund model calls through its free or Pro plan. Pro is sold monthly or annually, with the annual option billed once a year at a lower monthly rate; the pricing page carries the current amounts. Each plan's monthly allowance is split into a basic and an advanced pool, and which pool a model spends from is decided by what that model costs to run, not by the vendor's own naming. For unlimited bring-your-own-key (BYOK) usage, connect one OpenRouter key or a custom endpoint; credentials are encrypted with AES-256-GCM before storage and used only for the configured service. In BYOK mode ChatRound adds no model markup — the connected provider bills you directly at its rates, and no allowance is consumed. ## Architecture - **Frontend**: React + Vite + Tailwind CSS, served as static assets from Cloudflare Workers - **Backend**: Cloudflare Workers (Hono framework) - **Real time**: WebSocket connections through Cloudflare Durable Objects, streaming each model's answer as it is generated - **Database**: Cloudflare D1 (SQLite) for users, conversations, and messages - **Storage**: Cloudflare R2 for attachments - **Auth**: Google OAuth 2.0 with JWT session tokens ## Links - [Website](https://chatround.ai/) - [Public AI model comparison examples](https://chatround.ai/examples) - [Short version of this document](https://chatround.ai/llms.txt)