GPT-5 Review: A Powerful but Mixed Launch

TL;DR

GPT-5 launched from OpenAI as their smartest, fastest, most useful model with built-in thinking, and it is available to all users including the free plan. It works as a unified system that uses a real-time router to switch between a fast model and a deeper GPT-5 thinking model. Despite strong benchmarks and coding demos, community results and access issues reveal real limitations.

Transcript

Yesterday, OpenAI dropped GPT5. Yep, no joke. That's real. We've been joking, memeing, and hyping it up since GPT4 launched. I think it's safe to say that no matter what happened, OpenAI was not going to be able to meet everyone's expectations fully. Now, I want to mention I also did a watch party live with you guys. Some of you might have been the... Read More

Key Insights

• GPT-5 is available to all users including the free plan, which the creator praises as the biggest win of the release since it puts the latest technology in everyone's hands rather than locking it behind a paywall.
• GPT-5 is a unified system that pairs a smart, efficient model for most questions with a deeper GPT-5 thinking model for harder problems, using a real-time router to decide which to use based on conversation type, complexity, tool needs, and explicit intent.
• The router responds to explicit user intent, so typing a phrase like 'think hard about this' in the prompt signals it to switch to the GPT-5 thinking model for a more considered answer.
• Paid tiers scale usage: Plus subscribers at $20 a month get more usage, while Pro subscribers get essentially unlimited GPT-5 access plus GPT-5 Pro extended reasoning for better answers.
• Thinking mode dramatically changes math performance: on the AIME 2025 benchmark GPT-5 thinking reaches 99.6% and GPT-5 Pro hits 100%, but without thinking the score drops all the way to 71.0%.
• OpenAI focused on writing, coding, and health, calling GPT-5 their strongest coding model to date with the biggest gains in complex front-end generation and debugging larger repositories.
• The demos, including a jumping ball runner game, pixel art, a typing speed race, and a drum simulator, were built in a single prompt but are cherry-picked, so they do not represent the full real-world picture.
• OpenAI claims significant advances in reducing hallucinations, improving instruction following, and minimizing sycophancy, which is when a model overly hypes up or validates the user.

Questions & Answers

Q: Is GPT-5 available on the free plan?

Yes, GPT-5 is available to all users, including those on the free plan. The creator calls this the biggest slam dunk of the entire release and praises OpenAI for it, arguing that giving access to the latest and greatest technology to everyone is the whole point of AI. Paid tiers add more: Plus subscribers at $20 a month get more usage, and Pro subscribers get essentially unlimited access plus GPT-5 Pro extended reasoning.

Q: How does the GPT-5 unified system and router work?

GPT-5 is a unified system combining a smart, efficient model that answers most questions with a deeper reasoning model called GPT-5 thinking for harder problems. A real-time router quickly decides which to use based on the conversation type, complexity, tool needs, and your explicit intent. For example, saying 'think hard about this' in the prompt tells the router to switch to GPT-5 thinking. Once usage limits are reached, a mini version of each model handles remaining queries.

Q: What is the difference between GPT-5 with and without thinking on benchmarks?

Thinking mode makes a large difference in performance. On the AIME 2025 math benchmark, GPT-5 thinking nearly saturates the test at 99.6%, and GPT-5 Pro reaches a full 100%. However, GPT-5 without thinking drops all the way down to 71.0%. This shows that if your response does not use the thinking model, results on math will be significantly worse, so the reasoning mode is critical for harder problems.

Q: How much does Spotter Studio cost in this deal?

The video's sponsor, Spotter Studio, is running a limited time summer deal offering a yearly membership for $99, which the creator says is about 80% off the original price. Spotter Studio is built for serious creators and uses AI to help with ideation by brainstorming niche ideas from real world data, prioritizing which idea will succeed in the YouTube algorithm, and packaging it with titles and thumbnails. It also offers outliers, keyword rankings, and a daily feed of ideas.

Q: What did OpenAI focus on improving in GPT-5?

OpenAI focused GPT-5's biggest improvements on three domains: writing, coding, and health. They call it their strongest coding model to date, with the biggest gains in complex front-end generation and debugging larger repositories. They also touted significant advances in reducing hallucinations, improving overall instruction following, and minimizing sycophancy, which is when a model overly hypes you up or validates you completely all the time rather than giving honest answers.

Q: Why does the creator say the GPT-5 launch was a mixed bag?

The creator says GPT-5 has strong benchmarks and impressive cherry-picked demos, but the community's benchmarks differ from the ones OpenAI shows on its blog, and there are documented cases of GPT-5 not behaving as OpenAI expected. He also had a rough user experience: he only had access to GPT-5 on his phone, and all older models like 4.1, 4.5, o3, and o4 mini were completely gone, removing his reliable fallbacks and preventing direct comparisons.

Q: What coding demos did OpenAI show for GPT-5?

OpenAI showed several programs that GPT-5 built in a single prompt, though the creator stresses they are cherry-picked. The demos included a simple jumping ball runner game that plays like Flappy Bird, a pixel art creator, a typing speed race that looks almost like a real typing website, a drum simulator, and a lofi visualizer. The creator notes the UI looks clean and operates smoothly, especially impressive for being done in one generation, showing GPT-5 is capable.

Q: How does GPT-5 compare to older models on writing and health prompts?

In OpenAI's cherry-picked examples, GPT-5 clearly writes a better poem than GPT-4o and does a better job following all the points in a witty wedding toast prompt, writing more content overall. For a cancer risk health prompt, o3 was more clinical and gave a somewhat broken chart, while GPT-5 was more emotional, more like a chatbot, and gave a better overall breakdown. On a harder detailed prompt, GPT-5 thinking produced a much more comprehensive and detailed response than o3.

Summary & Key Takeaways

• OpenAI released GPT-5, branding it their smartest, fastest, most useful model yet with built-in thinking and state-of-the-art performance across coding, math, writing, health, and visual perception. It is available to every user including the free plan, a move the creator singles out as the release's biggest slam dunk.

• GPT-5 operates as a unified system: a fast efficient model handles most queries while a GPT-5 thinking model tackles harder problems, with a real-time router choosing between them. When usage limits are hit, mini versions take over, and OpenAI plans to merge all capabilities into one model soon.

• Benchmarks and cherry-picked demos look strong, with GPT-5 outperforming GPT-4o and o3 on writing and health prompts, but community results differ from OpenAI's numbers. Access problems, such as older models disappearing, and documented misbehavior show the launch was a mixed bag.