GPT-5: Not the AGI Messiah, but still pretty impressive

Want to read more write-ups like this? Subscribe to the Product Hunt weekly newsletter for a new one every Sunday 🌟

OpenAI just dropped GPT-5, and the hype train roared straight into AGI territory. In a live-streamed event that felt ripped from an Apple keynote, Sam Altman strode onstage, threw real-time coding challenges and puzzles at GPT-5 under spotlights, and even took audience Q&A. “This feels like talking to a PhD-level expert,” he quipped, as the hype engines revved up proclaiming AGI’s arrival.

In reality, GPT-5 is a massive leap in speed, accuracy, and flexibility, with a huge focus on AI coding assistance, but it remains a static snapshot trained through April 2025, not a self-evolving mind. I'm sorry to those of you who want your own WALL-E or Baymax, but AGI isn't here yet.

Here is everything you need to know, from the model lineup and ChatGPT’s six headline upgrades to vibe coding magic, benchmark records, and more.

The models you get

As has become the trend in AI model launches, GPT-5 is no single monolith. OpenAI offers five distinct engines for users and developers, so you pay only for the power and latency you need:

GPT-5

• The flagship, unified system that automatically routes simple prompts down a fast path and heftier queries into its deep-thinking core. Priced at $1.25 per 1 M input tokens and $10 per 1 M output tokens.

GPT-5 Thinking

• Invoke by typing “think hard” or toggling it in Plus, Pro, or Team. This engages the deeper-reasoning model for multi-step logic and complex problem solving. Same pricing as GPT-5.

GPT-Thinking Pro

• Available to Pro and Team subscribers. Extends context windows up to 128 K tokens for marathon codebases, legal analyses, or in-depth research. Same base rate plus surcharges for extra context .

GPT-5 mini

• A budget-friendly workhorse delivering roughly 80 percent of flagship performance at lower cost and latency. Priced at $0.25 per 1 M input tokens and $2 per 1 M output tokens .

GPT-5 nano

• The penny-pincher’s dream for massive-scale classification or trigger-phrase detection. Micro-tasks run at $0.05 per 1 M input tokens and $0.40 per 1 M output tokens, with micro-millisecond latency.

Six ChatGPT upgrades you’ll actually notice

Alongside the big model upgrades, the interface got some noticeable upgrades. Every ChatGPT user now benefits from GPT-5’s underlying power plus six tangible improvements:

Unified engine

The old model picker is gone. ChatGPT transparently routes your query to the optimal path, with an optional “GPT-5 Thinking” toggle on paid tiers .

Four personalities

Choose the tone that matches your mood:

• Cynic delivers dry, sardonic wit

• Robot offers precise answers with zero fluff

• Listener provides warm, empathetic reflections

• Nerd brings playful curiosity and enthusiastic exclamation points

UI customization

Accent-color controls let you theme chat bubbles, voice buttons, and highlighted text so your AI feels like part of your brand or matches your mood .

Advanced voice mode

Standard voice is retired. Paid users unlock near-unlimited speech with tweakable styles; free users gain bonus hours. Works seamlessly in custom GPTs .

Gmail and Calendar integration

Pro subscribers can link Google accounts so ChatGPT flags unread emails, plans your day, and even nudges you about that dentist appointment you keep postponing .

Safety and honesty boosts

Over 5 000 hours of safety testing have slashed hallucinations and taught GPT-5 to more readily admit “I don’t know” .

Better vibe coding capabilities

GPT-5’s vibe coding abilities kind of just hit warp speed or pretty close to it. Describe your dream web app in plain English–layout, colors, features, and within seconds you get a live Canvas prototype: HTML, CSS, JavaScript, hosting, and an interactive front-and back-end demo.

What once required copy-pasting snippets and manual setup now springs from a single prompt, letting anyone prototype full-stack web apps in minutes. This is getting the bulk of attention online, with vibe-coders spinning up apps like it's child's play. I even saw one person spin up a 3D game faster than I could hit the like button.

Benchmarks, benchmarks, benchmarks

Benchmarks should always be taken with a grain of salt. They are effectively a snapshot of a models capabilities under near perfect conditions. Sort of like the Big Mac you see on the ads vs the Big Mac you get in the bag. It gives you a good idea, but they're far from a perfect measure of real world usage.

• Math (AIME 2025, no tools): 94.6 percent

• Real-world coding (SWE-bench Verified): 74.9 percent

• Multilingual programming (Aider Polyglot): 88.0 percent

• Multimodal understanding (MMMU): 84.2 percent

• Medical reasoning (HealthBench Hard): 46.2 percent

• Graduate-level logic (GPQA without tools, via GPT-Thinking Pro): 88.4 percent

In production, GPT-5 Thinking cuts hallucinations by 45 percent versus GPT-4o and by 80 percent versus OpenAI o3 on complex prompts. On open-ended factual tests like LongFact and FActScore, it hallucinates six times less when reasoning is engaged.

What next?

It's become sort of a meme that when OpenAI drop their new shiny model, so too do the others in the space. With the launch of GPT-5, I wouldn't be surprised to see Google dropping a new Gemini model, Anthropic announcing a more powerful Claude, and Meta giving us an even bigger open-source model in Llama.

So I wouldn't take your eyes off the space yet, new models are probably coming en masse, and if the trend is anything to go by, vibe-coding, and voice interfaces are going to benefit immensely from these.

Not AGI, but still a leap

GPT-5 is a giant leap forward–faster, smarter, more flexible than any model before. Its tiered engines ensure you pay only for the power you need. Its personalities and UI flair make every chat feel custom. Its vibe-coding magic turns prototyping into child’s play.

Yet at its core, GPT-5 remains a human-trained snapshot, not a self-evolving mind. So go ahead–pick your tier, explore the new personas, and revel in those benchmark gains. If you were expecting AGI’s grand debut, you might want to keep your party hats on ice a bit longer.

P.S. Yes, it still loves em-dashes and I still want to vomit when I see one.