TL;DR
- GPT‑5 is the new default in ChatGPT, and it decides when to answer fast vs. when to “think hard” so you get expert‑level help without fiddling with model settings. You can even say “think hard about this” to trigger deeper reasoning. OpenAI
- It’s more accurate than prior models on real‑world questions and much less likely to hallucinate than GPT‑4o—and even more so when it’s in “thinking” mode. OpenAI
- Real upgrades you’ll feel: cleaner writing, steadier coding, and more cautious, useful health guidance (still not medical advice). OpenAI
- Study Mode (a guided, step‑by‑step way to learn) and automatic Gmail/Calendar/Contacts connectors make ChatGPT far more helpful for school and work. OpenAIOpenAI Help Center
- Access: Free users get GPT‑5 with limits, Plus raises those limits, and Pro ($200/mo) adds GPT‑5 Pro for the toughest tasks. ChatGPT
Why GPT‑5 “feels” like a PhD
Talking to GPT‑4 could be great—but you sometimes had to pick the “right” model or re‑prompt to get depth. GPT‑5 routes that choice for you. It runs as a unified system with a fast default model, a deeper reasoning model (“GPT‑5 thinking”), and a router that decides which one to use based on your prompt and context. If you want to force depth, type “think hard about this.” That’s why it often feels like a seasoned expert who knows when to take notes, sketch a plan, or pull in citations—without you micromanaging it. OpenAI
What changed in practice
1) Better answers with less hand‑holding
OpenAI reports big drops in factual mistakes vs. GPT‑4o, and larger gains when GPT‑5 is “thinking.” In other words, you’ll spend less time double‑checking basic facts and more time deciding what to do with the answer. OpenAI
2) Writing that lands
Expect stronger structure, clearer tone control, and more helpful follow‑ups. It’s easier to turn rough notes into a clean memo or a web page that sounds like you. OpenAI shaped GPT‑5 specifically to improve writing, coding, and health—the three things people do most with ChatGPT. OpenAI
3) Coding that holds up
GPT‑5 is better at front‑end generation and debugging across larger repos. If you’ve ever asked for a single‑file prototype or a quick test suite, you’ll notice steadier, more runnable results. OpenAI
4) Health answers with guardrails
It acts more like a careful thought partner—flagging concerns, asking context questions, and organizing what to ask your clinician. It’s not a doctor, but it’s better at helping you prepare for one. OpenAI
The “student who became a tutor”: new features that change the feel
- Study Mode turns ChatGPT into a step‑by‑step teacher instead of a quick‑answer machine. It nudges you with explanations, checks your understanding, and adapts as you go—useful for math proofs, exam prep, or learning a tool at work. OpenAI
- Automatic connectors bring your Gmail, Google Calendar, and Google Contacts into the chat (only if you enable them). Ask “Draft a reply to the calendar invite I just got” and GPT‑5 can reference the right thread or event automatically. You stay in one conversation; it fetches the context. OpenAI Help Center
Why it feels more trustworthy
OpenAI says GPT‑5’s answers are ~45% less likely to contain a factual error than GPT‑4o on web‑enabled queries, and when it’s thinking, ~80% less likely than OpenAI o3 on similar tests. It also shows fewer “agree‑with‑you” tendencies (sycophancy) and is trained to be honest about limits—for example, when the task is impossible or under‑specified. That means fewer confidently‑wrong answers and more “here’s what’s missing.” OpenAI
“PhD‑level” under the hood (for the curious)
If you care about benchmarks, GPT‑5 reports state‑of‑the‑art results across math (AIME 2025), real‑world coding (SWE‑bench Verified), multimodal understanding (MMMU), and health (HealthBench Hard). Benchmarks aren’t the real world, but they do mirror the steadier feel in day‑to‑day use. OpenAI
How to get GPT‑5 to do its best work for you
- Say what “good” looks like. Add audience, length, tone, deadline, and constraints.
- Invite deeper reasoning. Start with: “Think hard about this. List assumptions, trade‑offs, and risks.” OpenAI
- Ask for alternatives. “Give me three options, a cost/benefit table, and a 1‑paragraph recommendation.”
- Use Study Mode for any topic you truly want to learn (not just complete). It will slow down and guide you. OpenAI
- Connect your tools (if you’re comfortable) so it can draft replies, schedule around conflicts, or surface the right contact automatically. OpenAI Help Center
Mini playbook: examples you can copy‑paste
1) Strategy
“Think hard about this. We’re choosing between three product directions. Build a 1‑page brief with: goals, key assumptions, 3 options, risks, rough costs, recommended choice, and the first two experiments to de‑risk it.” OpenAI
2) Writing
“Here’s my messy outline + audience. Draft a warm, authoritative 800‑word post. Keep my voice. Ask me 2 clarifying questions before you write.”
3) Coding
“Generate a single‑file web app that [does X]. Use accessible HTML, responsive CSS, and inline comments. Add a quick ‘Getting Started’ note and one simple test.” OpenAI
4) Health prep
“I have an appointment about [topic]. Summarize key questions to ask, warning signs to mention, and a short context blurb I can share with my clinician.” (Not medical advice.) OpenAI
5) Learning
“Turn this syllabus into a 2‑week Study Mode plan with daily checkpoints, spaced‑repetition prompts, and a Friday self‑quiz.”
