How to Prompt Google Gemini 4 Argon

Gemini 4 Argon reasons before it replies. Skip the tricks and long role-play. Bring a clear goal, the right context, firm constraints, and a way to check the answer.

Tell Argon what done looks like, not how it should think.

Goal

One or two sentences on what you want

Context

Only the files, data, and facts that matter

Constraints

Rules, limits, deadlines, and anything to avoid

Output

The exact format, length, and sections

Verification

How it should check its own work, such as citing pages or running tests

Goal: find every renewal risk across these 12 supplier contracts. Context: the attached contracts plus our policy document. Constraints: rely only on the documents; flag anything unclear rather than guessing. Output: a table with supplier, renewal date, notice period, price change, risk level, and page reference, followed by a five-point summary. Verification: cite a page for every row and list any contract you could not fully read.

• Place large, fixed material such as documents at the top of the prompt
• Place your specific question below that material
• Request page or section references
• Ask it to say so when the documents do not contain the answer

• Switch on tasks mode for long, multi-step work
• Name the connected apps it may use
• State what needs your approval before it acts
• Request progress updates

LOW

Classification, summarization, and data extraction

MEDIUM

The default. General coding, tool use, document Q&A, and understanding video

HIGH

Multi-file software engineering, formal proofs, architecture design, and hunting for vulnerabilities

Choose the level in the API configuration, not in the prompt text. Try your task at all three levels.

• Use thinking_level, because thinking_budget returns an error
• Temperature, top_p, and top_k are ignored, so do not depend on them
• Supply a response schema for structured output
• Output is capped at 65,536 tokens per streaming request by default
• Reserve long output for answers that truly need it, such as hard proofs or large coding tasks

Argon automatically caches shared prompt openings of 6,144 tokens or more. Keep repeated instructions and documents at the start of each request, and put the part that changes at the end.

• Describe every tool clearly
• Favor one tool call at a time in multi-step loops
• Return tool results in the order the calls were made
• Have it state its plan before acting
• Keep humans in charge of risky actions

• “Before you answer, list what you would need to check.”
• “Give your answer, then your confidence and what could prove it wrong.”
• “Compare these options in a table, then recommend one and explain why.”
• “Write tests first, then the code, then run those tests.”

• Pasting everything you have rather than what matters
• Setting vague goals like “make this better”
• Requesting long outputs for simple tasks
• Running HIGH thinking on quick jobs
• Skipping verification