Big news from OpenAI this week! The company released not only its long-awaited GPT 5 model, but also two semi-open source models called gpt-oss-120b and gpt-oss-20b. (Perhaps OpenAI should ask ChatGPT for help with naming their models. LOL)
First GPT-5 news…
OpenAI calls GPT-5 their “smartest, fastest, and most useful model yet, with thinking built in” and claims that it’s better at coding, writing, and providing more useful health answers, and that it’s safer and more accurate. It’s now available to all users free and premium alike.
Updates for users include:
- Personalities – select from Cynic, Robot, Listener, or Nerd for personalities ranging from efficient and emotionless to playful and curious.
- Accent Colors – you no longer have to deal with ChatGPT in that sterile gray environment all day.
- Voice improvements – it understands your instructions better and allows you to customize its speaking style.
- Study mode – for step-by-step help learning a new subject.
- Google integration – the ability to connect your Gmail and Google Calendar
Users weren’t thrilled with the release, mostly because unlike in previous GPT model releases, OpenAI depreciated all of its other models with the launch of GPT-5 including 4o, 4.1, 4.5, 4.1-mini, o4-mini, o4-mini-high, o3, and o3-pro. During a Reddit AMA with Sam Altman on Friday, many users complained that GPT-5 wasn’t working as well as 4o did and requested that they bring it back.
Altman explained that the reason GPT-5 seemed “dumber” was because the router wasn’t working properly when it rolled out on Thursday, but now it’ll seem smarter moving forward. However the requests to bring back 4o were so plentiful and persistent that Altman said he would look into it. However realistically people will just move on from 4o, and OpenAI will continue to improve GPT-5 based on user feedback.
Personally, I think GPT-5 works just as good or better as 4o, which couldn’t seem to remember an instruction to save its life. What’s been your experience with GPT-5 so far?
Now onto gpt-oss-120b and gpt-oss-20b news…
First it’s important to note that the models are only semi-open, which means developers do not get the source code or training data.
The gpt-oss models come in two sizes: 120 billion and 20 billion parameters, which relate to how capable they are (higher is better). Both models are capable of running on the user’s own computer while the smaller one, which is text-only, can run on a smartphone — which is an incredible feat!
The last time OpenAI released an open model was in 2019, with GPT-2, which was fully open sourced in stages over several months. OpenAI didn’t say whether gpt-oss will eventually be fully open-sourced as well.
In the meantime, the lack of ability to view the data that went into the model could prevent financial services, healthcare companies, and other highly regulated industries from adopting them because they lack the transparency needed for compliance purposes. However wouldn’t they likely use OpenAI’s non-open models anyway? Perhaps that’s the goal.
In other AI news this week…
- OpenAI reached an agreement with the U.S. General Services Administration to offer ChatGPT Enterprise for just $1 per agency next year, so effectively free, leading Google and Anthropic to be like “WTF man?!?”
- The United Arab Emirates began offering its advanced Falcon AI programs completely free to position itself as a trusted alternative for developing nations. Rest of World wrote, “By offering free and open-source alternatives, the UAE is turning its petroleum profits into digital influence, and could become the go-between in a technology world split between the U.S. and China.”
