🔎 Preview: In today's issue
- ✨ Reasoning vs None-Reasoning models
- 🚀 GPT-5 and why it's interesting
- 📺 The story of building a six-figure AI SaaS
Hello and Happy Tuesday to the regulars and the 283 new subscribers 👋
We're back in your inbox to cover the latest launches and new features to help you ride that AI wave 🏄🌊 Hot off the back of the controversial GPT-5 release we've broken down some of the features of the latest model, shared a few upcoming events and a fireside chat with an AI SaaS founder to give you some ideas on what to build.
Let's jump in...
Explain Like I'm 5
Reasoning vs None-Reasoning models...and Why Does GPT-5 Matter?
OpenAI just dropped GPT-5, and everyone’s asking: is this the leap to “true” AI Not quite—but it’s a major step in how models think.
Let’s break it down:
- Reasoning Models try to solve problems step-by-step, like a human working through a tough math problem or planning a project.
- Non-Reasoning Models respond instantly, perfect for quick tasks like drafting an email or checking grammar.
Easy explainer
So, what’s new with GPT-5?
- Instead of picking a model for each task (chat, coding, analysis), GPT-5 auto-selects the right approach, reasoning or non-reasoning, based on your intent.
- It’s more honest: if it doesn’t know, it says so.
- It's noticeably less sycophantic: Model is less likely to be overly agreeable, thanks to post-training adjustments using real production data.
- It can remember way more (400k tokens!), making it better at handling long documents or conversations.
- it has multiple personalities that you can access. (This was tricky to find as it's only available on browser or mobile app and I'm a Mac app user).
- Gmail Google Calendar Integration: Connect directly for scheduling, email drafting, and active day management. This could be particularly interesting to avoid situations like below.
Avoid this
- It's competitively priced. Let's take running a review of 5 research papers. Assume each PDF is 10 K tokens and you want a 2 K-token structured summary. Total input: 50 K tokens (0.05 M). Total output: 10 K tokens (0.01 M)
- GPT-5: $0.1625 ( $0.0625 input + $0.10 output )
- GPT-4o: $0.225
- Claude Opus 4.1: $1.50
It's 10% of the cost of Claude Opus 4.1 with the similar results. This should have dramatic effect on how we choose models.
- API access offers three models: Regular, Mini, and Nano (Nano being the cheapest) each can be run at four reasoning levels (minimal, low, medium, high). So more granularity compared to the ChatGPT which only has the 2: thinking or fast mode
**🧠 Why does this matter?**
You get the right kind of “thinking” for the job. Fast answers when you need them, deeper analysis when it counts.
For all it's wooden acting their product launch video was packed with innovative product improvements.
🤔 What else was interesting?
It's hard to please everyone; so the internet has been keen to express their opinions. Here's a few
- it's dry
James Wang
[
GPT-5 is just no fun to talk to 😕
](https://x.com/draecomino/status/1954830526201926113)
10:1 AM • Aug 11, 2025
- it thinks it's too smart
@levelsio
[
I hate ChatGPT 5, it's so bad, it's so lazy and it won't let me switch back to 4o cause I'm on Plus, this might really make me switch to Anthropic's app now, I'm actually annoyed by how bad it is, it's making my productivity go 10x lower cause nothing it says works
](https://x.com/levelsio/status/1954208088065556676)
4:48 PM • Aug 9, 2025
- it still does things a human intuitively does
Andrej Karpathy
[
@KocKaanKarakas Two situations: 1 I wave a co-worker to my monitor to show them a file I have open. "Is this right" 2 I sit down someone at a table. They have 2 hours to respond. This is an exam. The stakes are high. "Is this right?" The humans collaborator knows 1 vs 2. The LLM doesn't know
Kaan Karakas
@KocKaanKarakas
@karpathy Yeah totally agree. The models are leaning too much into overthinking, which is great for deep complex stuff but slows things down for quick checks. A built in way to set the depth, from quick glance to full deep dive, would make things way smoother instead of having to say it
](https://x.com/karpathy/status/1954231225230790716)
6:20 PM • Aug 9, 2025
What do you think of GPT-5 Hit or miss
🔊 On your radar
Keep your finger on the pulse of what we think is hot in the space right now 🔥
⭐️ OpenAI drops first open-weights e.g. open source reasoning models
Oh wait...they did something else this week
👉 Why it matters: Two models (20B/120B params) built for tool calls and agents, but crawls at 10 words/min according to Ben Tossell. Fine-tuning potential massive (eg. think highly regulated industries that can now build out models with a known LLM). We think it's a small snipe at Meta after their little tit for tat recently.
WTF is Open Weight See the knowledge shot in the subscriber section for an explainer.
⭐️ Google Genie 3 creates persistent game worlds. This is fun
👉 Why it matters: This feels far away from our day to day however it generates realistic environments you can explore with a 1-minute memory. It's early doors but we creating worlds like this required a huge amount of designers and developers; now it's a simple text prompt.
Angry Tom
[
We're so cooked. This is Genie 3, the most advanced world simulator ever created. Given a text prompt, it can generate dynamic worlds that you can navigate in real time. 10 wild examples:
video
](https://x.com/AngryTomtweets/status/1952785519773176198)
6:35 PM • Aug 5, 2025
⭐️ Framer Launches On Page Editing
👉 Why it matters: Handing over a project has always been tricky for agencies and freelancers but Framer just made the process easier with on page editing. This means that clients can click and make changes to copy, components and images directly from the live site.
[
](https://api.filekitcdn.com/e/vgxjtp9FsrVP46eWzFo3R3/8x4Znbm7oQ2huPJ9sp3HY3/player)
With Contra releasing its Indy product last week it's expected that this unlocks millions in more opportunties for Framer creators.
🔗 Bonus links for subscribers
- Lovable shipped just hosted their cohort's first Demo Day. Discover what the top 10 finalists created using the platform:
- MasterInbox just took home Paddle's AI Launchpad Demo Day for $10k: helping lead gen agencies never miss a reply
- Google Jules (an asynchronous coding agent) is out of beta and available for you to try to help fix bugs and run tests
- Claude can now reference your past chats. Plus: a Claude Code one-pager for prompt engineering and a 2 hour course on Claude Code
- Make and other tools are rolling out GPT-5 Support including Bolt, Lovable and more
☕️ Knowledge shot — Open Weight refers to AI models where the actual trained parameters (the "weights") are freely downloadable, but the training code might not be. Open Source means everything is available—weights, code, training data, the lot. OpenAI's new gpt-oss models are open weight—you can download and run them locally, but you can't see exactly how they were trained.
📺 Building a six-figure AI SaaS with no-code with Alex Rainey and Tom Wesolowski
Alex Rainey, co-founder of My AskAI, shares his journey from corporate consulting to building a $40k MRR AI customer support company using 75% no-code solutions.
After raising $1.5m for his first startup and learning hard lessons from pivots and exits, Alex will reveal how he leveraged no-code tools to rapidly iterate, validate, and scale his current venture without external funding. Discover the practical strategies and tool stack that enabled a lean team to build a profitable AI SaaS competing with heavily-funded competitors.
He's also got a new mini product for Bubble builders. DM him and check it out
▶️ Watch Video
🗓️ Events to Take Note Of
🇬🇧 Want to work on something IRL with other AI creators Join the inaugural About Product, About AI Hack Day in London this Saturday at Ramen Space. Just want to watch cool talks The next AI Tinkerers meetup London is the 28th of August
🇺🇸 Bubble meetups in San Fransisco (today!) and in New York City Tuesday August 19th
🌎 International Generalist Day is back and hosting events across the globe on September 3rd powered by Lovable
🌐 The Oscars of AI film, music and games just dropped: flex your skills to take home the ultimate prize in November! Introducing The Chroma Awards - expect free tool trials starting from Sept 1st
🖱️ Discover more events on the new Create With events directory!
💭 Final Thoughts
Can you spot the issue with this graph 🔍
It transpires that it was human error, late into the night, that led this graph to not being verified. So don't be too hard on yourself the next time you have a typo on a slide, especially during summer - it even happens to the OpenAI team.
See you next week!
With good vibes ✨
{% poll 28123 %}{% for option in poll.options %}{% endfor %}
{{ poll.question }}
[
]({{ option.url }})
[{{ option.value }}]({{ option.url }})
{% endpoll %}
{{ address }}