Prefers short over persuasive. Gemini 3 answers direct instructions better than elaborate ones, and puts your question last when there is a lot of data.
How to prompt it
Be brief and literal. State the goal, skip the throat-clearing.
Use consistent delimiters between sections — XML-style tags or Markdown headings both work.
With a long document or dataset, put the data first and your instruction at the very end.
Define any term that could be read two ways. Gemini takes ambiguous parameters literally.
For a conversational voice, ask for it. The default is efficient, not chatty.
Raise thinking_level rather than writing "reason step by step" into the prompt.
Strengths
- Strong multimodal handling — text, image, audio and video in one prompt.
- An explicit thinking_level control from minimal to high, so reasoning depth is a setting rather than a prompt trick.
- Efficient by default: concise answers without being asked.
- Handles very large context, with documented advice for arranging long inputs.
Weaknesses
- That same terseness reads as curt in customer-facing writing unless you explicitly ask for a warmer register.
- Persuasive or emphatic prompt language ("this is extremely important!") adds nothing and Google advises against it.
- Old chain-of-thought scaffolding is now counterproductive: Google recommends simplifying the prompt and raising thinking_level instead.
Models
Google’s model overview does not publish context and output limits in one table; open the individual model page for current figures.
Gemini 3.7 Flash
Built for complex coding, agentic workflows and reliable multi-step execution.
- Context
- —
- Max output
- —
Good for: Agentic and coding work where you want speed alongside multi-step reliability.
Watch out: Terse by default; ask explicitly for explanation alongside output.
Gemini 3.6 Flash
Balances speed and multimodal capability for general agentic tasks.
- Context
- —
- Max output
- —
Good for: Mixed text-and-image work at interactive speed.
Watch out: Set thinking_level deliberately for anything analytical.
Gemini 3.5 Flash-Lite
The fastest, most cost-effective model in its generation for high-throughput work.
- Context
- —
- Max output
- —
Good for: Classification, tagging, and very high request volumes.
Watch out: Needs tighter, more explicit instructions than the larger tiers.
Gemini 2.5 Pro
Previous-generation flagship for complex reasoning and coding.
- Context
- —
- Max output
- —
Good for: Workloads already tuned against it, and cases where you want the older, more verbose default.
Watch out: Prompts written for 2.5 usually need trimming before they work well on the 3.x line.
Sources and grounding
- 1Gemini 3 developer guideGoogle (Gemini API docs)Source for the brevity, delimiter, instruction-last and thinking_level guidance.
- 2Prompt design strategiesGoogle (Gemini API docs)General prompt design reference, including few-shot guidance.
- 3ModelsGoogle (Gemini API docs)Live model list and per-model specification pages.
- 4Google Prompting EssentialsGoogle (via Coursera)Google’s own end-user course, home of the TCREI framework.