AI Catch-Up: GPT-5 vs. Everything Else

Round-up of the major AI releases of the past month.

What’s up, fellow AI nerds and nerdettes?

I’m back in action after a relaxing summer break with my family.

Unfortunately, AI labs kept releasing new stuff in the meantime, despite my desperate letters begging them to hold off until I returned.

This means we have a month's worth of Sunday Rundowns to catch up on.

So let’s take stock of all the noteworthy AI releases since mid-July.

Buckle up!

🧠 The big one: GPT-5 saga

On August 7, OpenAI announced its long-awaited and much-hyped GPT-5:

On the surface, there was a lot to like about GPT-5:

• The underlying reasoning model was the best one yet, topping LM Arena and Artificial Analysis leaderboards across every benchmark…and it was even available to free ChatGPT accounts (with capped usage).

• The GPT-5 models were also faster and much cheaper than comparable API offerings from other frontier labs.

• The convoluted ChatGPT model picker was gone, replaced by a smart router that automatically picks which model to use and how much reasoning effort to dedicate to the task based on the context.

But within hours of the livestream, things began to unravel.

Soon, you could find strong opinions on either end of the spectrum, from classic “This changes EVERYTHING!” hype to “This sucks ASS!” outrage and everything in between.

On the one hand, GPT-5 had many admirers when it came to tricky coding challenges, where it seemingly excelled at one-shotting solutions where other models struggled.

On the other hand, plenty of observers were disappointed by the incremental rather than revolutionary advancements. GPT-5 wasn’t a massive leap towards AGI some had expected it to be. Notable LLM skeptic Gary Marcus celebrated “Gary Marcus Day,” feeling vindicated by GPT-5’s underwhelming improvements, persistent hallucinations, and other shortcomings.

On Reddit, regular users were up in arms because GPT-5 didn’t pass the vibe check compared to their beloved GPT-4o, which was now gone. Trending threads ranged from “Bring back ChatGPT-4o” all the way to “Please bring back GPT-4o.”

The pressure was so strong that OpenAI caved almost instantly, bringing GPT-4o back for paid users within 24 hours.

So in a matter of days, we went from the old, messy model picker:

…to the clean GPT-5 selector…

…to whatever the hell is happening here:

In my own brief interactions so far, GPT-5 was a mixed bag.

On several in-depth tasks, it seemed to perform on par with my prior go-to o3 model. Yet I wasn’t immediately blown away or noticed any stand-out leaps.

When putting together this summary, I came across the “Jumping Ball Runner” game coded by GPT-5 on OpenAI’s official announcement page.¹

See if you can spot a teeny-tiny issue with it—pay close attention, it’s subtle:

Ah! So I lose the game when acute carpal tunnel syndrome eventually kicks in? Clever!

But I don’t want to mock GPT-5 too hard here. It’s a solid model, even if hallucinations and other LLM issues haven’t magically disappeared.

The only way you’ll know if GPT-5 is right for you is by testing it out.

Feel free to share your experiences in the comments. I’m curious to hear what everyone thinks!

In case you just can’t get enough of the GPT-5 backstory, I compiled a NotebookLM notebook with a spread of positive and negative coverage, complete with a mind map, FAQ, and audio/video overviews:

Just log in with your Google account and chat with the sources at will:

🗞️ AI news you might’ve missed

And now, for the classic Sunday Rundown-style quick-fire catch-up:

• Adobe added better scene controls, the ability to generate sound effects, and other improvements to Firefly Video.

• Alibaba Cloud released Qwen3‑235B‑A22B, a reasoning model that can seamlessly switch between thinking and non-thinking modes.

• Amazon  launched Kiro, an AI‑powered IDE that helps developers go from concept to production by turning prompts into detailed specs.

• Anthropic news:

  • The updated Claude Opus 4.1 is better at coding, reasoning, and agentic tasks. (Available to paid users.)

  • Claude Code has been upgraded with automated security reviews that scan pull requests and flag risky tweaks.

• The Browser Company rolled out a $20/month Dia Pro plan with unlimited AI chat and “skills” for its AI-powered Dia browser.

• Cursor released BugBot, an assistant that analyzes code changes to proactively spot bugs or security issues.

• Decart AI launched MirageLSD, a diffusion model that can restyle any live video footage on the fly. (Try it for free.)

• ElevenLabs now has a music-making tool called Eleven Music that creates full tracks from text prompts, cleared for commercial use. (Try it for free.)²

• Genspark news:

  • AI Developer is an autonomous coding agent that plans, codes, tests, and ships complete apps on your behalf.

  • AI Meeting Notes is a smart recorder that can capture meetings via Apple Watch and provides instant summaries with action items.

• Google news…just so much stuff:

  • AI-powered calling in Search can call local shops to check pricing, availability, etc. on your behalf. Ongoing limited rollout in the US.

  • AlphaEarth Foundations is a virtual satellite model that maps our planet with incredible precision and combines dozens of information sources.

  • The new Create tab in Google Photos lets you restyle your photos or turn them into short video clips. (Rolling out in the US only for now.)

  • Deep Think is an advanced reasoning mode for Gemini Ultra subscribers that can solve complex problems using parallel thinking techniques.

  • Genie 3 is an experimental-stage world model that can generate real-time, interactive, persistent 3D environments from a text prompt.

  • An ongoing AI-powered Google Finance experiment can answer complex questions and provide complex charts based on real-time data and news.

  • Guided Learning mode in Gemini can break down any topic into step-by-step learning guides with visuals, quizzes, and other study aids.

  • NotebookLM now has Video Overviews that turn your sources into an animated slide deck with a descriptive voiceover.

  • Opal is a no-code tool that turns prompts into shareable apps you can edit in an intuitive visual editor. (Try it for free. US only, so you’ll need a VPN.)

  • Storybooks feature in Gemini can create complete storybooks with visuals and narration from simple prompts or your own uploaded images.

  • Web Guide is a Search Labs experiment that uses AI to group search results into topic clusters to help structure the information.

• Higgsfield dropped an Image Reference Tool + Browser Extension that lets you right-click any image to recreate it or restyle it without prompting.

• Hugging Face rolled out AI Sheets, a no-code spreadsheet tool that lets you build, enrich, and transform datasets. (Try it for free.)

• Hume AI’s EVI 3 model can clone your voice while also mimicking your style, tone, etc. (Try the free demo).³

• Ideogram launched Ideogram Character which can generate consistent characters in any style from a single reference image. (Try it for free.)

• Microsoft news:

  • Copilot Mode in Microsoft Edge can see your tabs, understand your tasks, and help you compare, decide, and act on information.

  • GPT-5 is now integrated into Microsoft’s offerings like Copilot, GitHub, and Azure for consumers, businesses, and developers.

• Mistral news:

  • Codestral 25.08 is a coding model for dev teams that writes, edits, and runs multi-step workflows directly inside your IDE.

  • Le Chat got lots of upgrades and can now perform deep research, chat via voice, edit images, and organize your work into Projects.

  • Voxtral is a family of affordable open‑source speech models that can understand, transcribe, and summarize audio input.

• Midjourney video model can now generate 720p HD video clips as well as handle start/end frames and video loops.

• OpenAI news:

  • Agent mode in ChatGPT can browse the web, code, schedule appointments, and handle other tasks using its own virtual computer.

  • ChatGPT has been revamped to reinforce healthy screen time habits and be a helpful guide in personal challenges.

  • gpt-oss-130b and gpt-oss-20b are open-weight reasoning models with top-tier performance that can be fine-tuned and run offline on consumer devices.

  • Record mode in the MacOS ChatGPT app captures voice notes or meetings to generate summaries that you can edit and work with.

  • Study mode turns ChatGPT into a tutor, guiding users with Socratic questions, personalized hints, and quizzes to help them absorb information.

• Proton launched Lumo, a privacy-first AI chat that requires no sign-ups, encrypts all chats, and doesn’t maintain logs or chat records.

• Runway news:

  • Act-Two is a motion capture tool that can animate a character from a driving video and a reference image or video.

  • Aleph lets you edit video footage using simple text prompts to change camera angles, lighting, styles, remove objects, and more.

• Tencent open-sourced HunyuanWorld 1.0, which turns text prompts or images into explorable 3D worlds.

• xAI rolled out its uncensored video model, Grok Imagine, which can generate 15-second video clips.

• Z.ai released an impressive open-weight model family called GLM-4.5 that matches many top closed models on agentic tasks and tool use.

🫵 Over to you…

Phew! I think that covers the major headlines, but do let me know if I’ve missed something.

Have you been keeping up with the AI news avalanche this summer, or have you taken a step back as I did? What’s been the most noteworthy AI release as far as you’re concerned?

Leave a comment or drop me a line at whytryai@substack.com.

---

Thanks for reading!

If you enjoy my writing, here’s how you can help:

• ❤️Like this post if it resonates with you.

• 🔄Share it to help others discover this newsletter.

• 🗣️Comment below—I love hearing your opinions.

Why Try AI is a passion project, and I’m grateful to those who help keep it going. If you’d like to support my work and unlock cool perks, consider a paid subscription:

1. It’s still there as of this writing, in case you want to take it for a spin yourself.

2. Time for Battle Of The Bands IV?

3. Damn, looks like I’ll also have to update my AI voice cloning test soon.