Chat with paper

Ask questions, pick models, and turn on thinking mode.

The right-hand chat panel in the Reader lets you ask anything about the paper you're reading. Behind the scenes we retrieve the most relevant chunks of the paper and pass them to the model as context, so answers are grounded in what the paper actually says.

Asking a question

Just type and press Enter. Some prompts to try:

• "Summarize the methodology in plain language."
• "What datasets are used and how were they constructed?"
• "Compare the results in Table 2 with the related work in Section 5."
• "Walk me through Equation 7 step by step."

Answers stream in token-by-token. You can stop generation at any time with the Stop button.

Citations

Inline chips like [1], [2] link back to the paper chunks the model used. Click a chip to jump to that chunk in the paper. This makes it easy to verify a claim or read the surrounding context.

Picking a model

Use the model picker above the chat input to swap between providers. Each provider has a fast small model and a slower flagship.

• Free users: GPT-5 Nano, Claude Haiku
• Pro users unlock Claude Opus, GPT 5 Pro models, Grok, Gemini, and more

Use the small fast models for one-off questions and the flagship models for deep reasoning, math, or multi-step analysis.

Thinking mode

Models that support extended reasoning (Claude Opus, GPT 5, Grok 4 reasoning, Gemini 3) show a Thinking toggle next to the model picker. Turning it on lets the model use a private scratchpad before answering. It's slower but noticeably better on hard questions.

Some models let you pick low / medium / high thinking effort. Start at low and bump up if you're not getting the depth you want.

Conversations

Each chat session is saved with the paper. You can:

• Switch between past conversations from the chat history menu
• Start a fresh thread without losing the old one

Conversation history lives on your account, so you'll see it whenever you reopen the paper.

Tips

• Quote a passage in your question to focus the model on a specific paragraph
• For long PDFs, ask one section at a time. The answer quality is much higher than asking the whole paper at once
• If a model refuses something benign, swap providers; they have different safety policies