Intro
GPT-5 released today! I watched the full 1 hour 15 minute release video and here are my takeaways:
Honestly, underwhelming. Given previous releases and the speed at which AI has been moving it was hard not to expect something completely game-changing coming out of the GPT-5 release. SamA has his own setup, like Jobs did with Apple, and the look and feel of the OpenAI keynote did a good job of resonating with those vibes, yet the product itself did not match the grandeur that often came along when the black turtleneck, blue jeans-wearing, Apple co-founder took the stage.
The biggest focus of the launch seemed to be based around coding capability. Time will tell how advanced this new coding system is, but there was high praise from the Cursor CEO & Co-Founder who not only shared the stage at times, but has begun implementing GPT-5 into his product. The other line OpenAI seemed intent on getting across was that of a “PHD in your pocket”. This is cool, but nothing all that new. Based on current ecosystem velocity, and SamA’s tweeting the night before, I almost expected a release that teased ASI, or even some AGI, yet GPT-5 falls short of both. By no means does this mean failure, just a bit underwhelming.
Here are the tangible positives:
The new release comes with a huge drop in hallucinations (20% with GPT-4o → 4.6% on GPT-5). Hallucinations have been a longstanding issue dating back to GPT-1, and eradicating them will significantly boost trust in daily use. Another standout was ChatGPT’s deeper integration into the Google ecosystem. It now supports access to Gmail and Google Calendar, kind of a surprising move given the competitive wall between OpenAI and Google.
The most jaw-dropping stat came from the industry-specific section of the presentation, where GPT-5’s agentic success rate for tasks within the Telecom industry reportedly jumped from 49% to 97% in just two months. If accurate, that suggests dramatic improvement in AI reliability for specific jobs currently done by humans. Finally, the coding agent demonstrated the ability to autonomously push code, identify errors, and rewrite on its own, a clear signal of where long-term agentic workflows are headed.
Other notable pieces of the release included:
GPT-5 will be available for free users.
Learning/education was spotlighted, but there didn’t seem like a ton of new features around it.
Wonder if this is related to usage charts reflecting heavier GPT usage while school is in session.
Medical capability was a big focus, but once-again nothing blew my mind feature-wise. They brought out an employee who was a cancer survivor; she discussed how she used GPT-5 to better understand her diagnosis. Cool use of the product, not game-changing for most.
Writing seems to be improved, that’s also something that will be easier to gauge over time. The Em-Dash or “ChatGPT-Dash” is still there, was wondering if they would try to nix it.
Voice chat also got upgrades. Longer limits on voice for all users, including advanced voice for free users. GPT-5 voice demoed the capability for stricter response parameters, such as one word answers - big fan of that! Often, I have to instruct it many times to not give long drawn out responses.
General chat updates include personality based GPT’s such as professional, strict, empathetic etc. Also included some UI features such as colored chats for pro users.
API users will have access to GPT-5 ($1.25), GPT-Mini ($0.25), and GPT-Nano ($0.05). Price per 1M tokens.
The one thing that I was really hoping to see was a major upgrade to the GPT-Agent, but we’re gonna have to wait on that. Overall, these are some good upgrades, but at the end of the day it seems more like a move that would take GPT 4.5 → 4.6 rather than all the way → GPT 5.
That’s all I got. If you saw anything you have input on, feel free to shoot me an email: [email protected].