Problem
Below are some of the most popular LLM models:
| Provider | Model Family | Chat (Standard/Instruction) | Reasoning (Thinking/Extended) | Primary Use Case |
|---|---|---|---|---|
| OpenAI | GPT-5.4 Series | GPT-5.4 / Pro | OpenAI o5 / o3 |
Frontier logic, scientific research, and 1M+ context tasks. |
| GPT-5.4 Mini/Nano | 5.4 mini / nano | o4-mini (API) |
Cost-efficient agents, mobile apps, and summarization. |
|
| Open Models | gpt-oss-20b | gpt-oss-120b |
Local deployment with SOTA reasoning capabilities. |
|
| Gemini 3.1 Pro | Gemini 3.1 Pro | Gemini 3.1 Pro (Thinking) |
Multimodal reasoning, enterprise "vibe coding," and heavy logic. |
|
| Gemini 3.1 Flash | 3.1 Flash / Flash-Lite | Gemini 3.1 Flash (Thinking) |
Real-time agents and low-latency production pipelines. |
|
| Gemini Live | Gemini Live (Flash) | Native Audio Thinking |
Human-like voice/video interaction with real-time logic. |
|
| Anthropic | Claude 4.6 Series | Opus / Sonnet 4.6 | Adaptive Thinking Mode |
Security audits, massive codebase refactoring, and ethics-aligned logic. |
| Claude 4.5 | Haiku 4.5 | Haiku 4.5 (Extended) |
High-throughput data processing with "thinking" steps. |
|
| DeepSeek | DeepSeek V3.2 | DeepSeek-V3.2 | V3.2-Speciale / Reasoner |
Competitive coding, STEM problem solving, and open-weight logic. |
| DeepSeek V4 | V4 Lite (Current) | Full V4 (Late April 2026) |
Future-state "Engram" memory and unified MoE reasoning. |
One of the biggest challenge faced by people now is: which one should I use?
Chat Model vs Reasoning Model
First of all, it's important to understand the key differences between Chat Models and Reasoning Models:
| Aspect | Chat Models | Reasoning Models |
|---|---|---|
| Primary Purpose | Designed for natural conversation and back-and-forth dialogue.
Think of it like talking to a knowledgeable friend who’s good at casual conversation. |
Built specifically for step-by-step problem solving and logical analysis.
Like having a methodical expert who breaks down complex problems. |
| Strengths | Great at small talk, general knowledge sharing, and maintaining context in conversations. They understand social cues and can adapt their tone. | Excel at math, coding, scientific analysis, and situations requiring careful logical thinking. They’re like having a detective who methodically solves puzzles. |
| Output Style | More conversational and fluid, often providing context and related information naturally. | More structured and analytical, often showing their work and explaining each step of their reasoning. |
| Best Use Cases | Customer service, general Q&A, writing assistance, brainstorming | Complex problem solving, mathematical analysis, scientific research, strategic planning |
In short,
Use Chat Model: for conversation and content creation
Use Reasoning Model: for math, coding, scientific analysis, and situations requiring careful logical thinking
When to Use What
Use the right tool for the right thing. Here’s a quick Cheat Sheet when to use what:
| If you need… | Then use… | Remarks |
|---|---|---|
| An all-purpose, fast, balanced AI (your go-to) | ChatGPT or Google Gemini | Your all-around, fastest, most versatile tool. Great for general conversation, quick answers, and understanding images. It’s the default “good at everything” choice. |
| To solve a very complex puzzle or logic problem; coding | Any reasoning model | Your specialized logic calculators. Use these when you need the AI to “show its work” and reason step-by-step without mistakes, like for advanced coding or intricate strategy. |
| Any PDFs more than 20 pages e.g. to analyze a huge document or book | Google Gemini / Google NotebookLM | Your super-powered document reader. If you need to ask deep questions about a whole book, a long contract, or a massive report, this one has the best memory. Also refer to this article. | Any research that demands facts | Google NotebookLM | Google NotebookLM answers based on your sources. To help you verify this, it provides inline citations that link directly to the text in your sources. | For research with tons of PDFs to digest | Google NotebookLM | For free users, you can upload up to 50 sources. For paid users, you can upload up to 300 sources. The sources can include documents (PDFs, Word, Powerpoint), images (JPG, PNG), audios (MP3), videos (YouTube links or MP4) and URLs. |
| To write an essay, blog post, or nuanced content | Claude Sonnet | Your thoughtful writer. It often has a slight edge in understanding nuance, tone, and producing high-quality, structured long-form content. |
| To learn a new topic | ChatGPT Study Mode | It uses the Socratic Learning style. If you enjoy learning new thing, Try this - so much more fun! |
| If you learn best by listening (e.g. podcast) | Google NotebookLM Audio Overview | Google NotebookLM has the mobile app for iOS and Android. So you can generate the podcast and listen to it on your mobile. |
| To search for latest info / Web Search / Google Search |
Perplexity | It’s like Google Search + ChatGPT – live search, no ads, summarize the results for you, with direct links to sources. |
| Corporate videos | Gemini Storybook or Google NotebookLM Video Overview | You can refer to this article for more details: From Bedtime Stories to Boardrooms: The Power of Google Gemini Storybook |
| Transcribe meeting minutes | Google Gemini / Google NotebookLM | You can upload the MP3 audio recordings to generate the full transcripts. |
| YouTube | Google Gemini | Google owns YouTube. It has direct access to all the YouTube videos and transcripts. Anything YouTube, just go with Google Gemini e.g. to produce summary, generate full transcript, etc. |
| Image generation / editing | Google Gemini or Google Studio AI - Nano Banana | For best results, use Google Gemini's Nano Banana 2. For norming image editing and stock photo generation, Google Studio AI's Nano Banana is good enough - it's still free! |
| OCR (Optical Character Recognition) | Google Gemini / Google NotebookLM / ChatGPT | Single page - you can go with Google Gemini or ChatGPT. For multiple pages, use Google NotebookLM - data tables |
| Unstructured data (e.g. images, PDFs, receipts, supplier invoice) to structured data | Google NotebookLM data tables | You can refer to this article for some sample prompts you can use: 06. Google NotebookLM Data Tables - tips and techniques. Once the data table is generated, output to Google Sheets first, then Export to Microsoft Excel. |
| Infographics | Google Gemini / Google NotebookLM | Use Google Gemini is you need to produce 4K resolution. Use Google NotebookLM if you are generating infographics from multiple sources. You can refer to this article to view the different styles of infographics you can produce and their corresponding prompts: 05. Google NotebookLM Infographics - tips and techniques |
| English / Chinese translation; Chinese homework; Anything involving Chinese | Deepseek | Its training data includes more Chinese |
Add comment