OpenAI released GPT-5 in April 2026. If you read the announcement, it was full of benchmark scores and technical jargon — "x% improvement on MMLU," "y parameters," "new architecture." For most people, those numbers mean nothing. Here is what actually changed for someone who just wants to use ChatGPT to get things done.
Better Writing, Less Editing
The most noticeable improvement is writing quality. GPT-5 produces text that reads like a competent human wrote it, not like an AI optimized a sequence of tokens. The giveaway phrases are gone — you will not see "in today's digital landscape" or "delve into" unless you specifically ask for that style. The model has a more natural rhythm, better sentence variety, and actually understands tone shifts. If you ask for "casual but professional," it delivers something that sounds like a real colleague.
In a blind test I ran — 20 emails, half written by GPT-4o and half by GPT-5, with the labels removed — 14 out of 20 people preferred the GPT-5 versions. The most common comment was "this one sounds like a person." That is the headline improvement.
Hallucination Is Down, Not Gone
OpenAI claims GPT-5 hallucinates 40% less than GPT-4o on factual questions. Independent testing mostly confirms this — Vectara's hallucination leaderboard shows GPT-5 at 2.1% hallucination rate compared to GPT-4o's 3.8%. That is meaningful progress. But let me be clear: it still makes things up. If you ask it about a specific court case from 2024, it might fabricate the details. You still cannot blindly trust facts from any LLM, GPT-5 included. The improvement is real but the fundamental limitation remains — it is a prediction engine, not a database.
The Memory Feature Finally Works
ChatGPT's memory feature was technically available before GPT-5, but it was unreliable. It would forget context after a few conversations or remember the wrong things. With GPT-5, memory actually works. It remembers that you prefer bullet points, that you have two cats named Luna and Simba, and that you work in marketing. You can view exactly what it remembers in settings and delete anything you do not want stored. This turns ChatGPT from a tool that treats every conversation as a fresh start into one that builds continuity.
For returning users, this changes everything. You no longer need to reintroduce yourself and your preferences every time. Just start typing.
Advanced Voice Mode Now Included
Voice conversations with ChatGPT used to be a Plus-only feature at $20/month. With GPT-5, the free tier gets limited voice access — about 15 minutes per day. The voice quality is stunningly good. It detects your tone, pauses naturally, and can even adjust its speaking style mid-conversation. I have found myself using it for brainstorming while cooking, dictating emails while walking, and practicing interview questions.
The Plus tier still gives you unlimited voice and access to the highest-quality voice model, but the free taste is generous enough to decide if you want to upgrade.
What Did Not Change
The free tier still has message limits — roughly 30 GPT-5 messages every 5 hours before it drops back to GPT-4o mini. Image generation is still limited on free. The API pricing increased slightly (about 15% per token), but subscription pricing stayed the same. So if you are a Plus subscriber, your $20/month still buys you everything. The upgrade was free.
Bottom line: GPT-5 is not a revolution. It is a meaningful refinement. The writing is better, the voice mode is usable, and memory finally works. If you were on the fence about whether ChatGPT was "good enough" — it just got better enough to tip the scale.