IN THE SPOTLIGHT
ChatGPT-5 is here and OpenAI goes (kinda) open

OpenAI just dropped ChatGPT-5, and it’s not just a slight upgrade. It’s faster, smarter, and can handle way more complexity, all while still being available for free to regular users. (Pro users get more firepower, of course.)

Here’s the scoop, minus the fluff:

1. It feels like a brainy team in your pocket.

GPT-5 writes better, codes faster, and handles real-world tasks like a pro. Think of it like having a bunch of PhDs you don’t have to manage.

2. Context overload? No problem.

It remembers more: up to 256,000 tokens (yep, a whole book’s worth). And rumor has it, future versions might handle up to a million.

3. Smarter, not just faster.

It reasons better. Whether it’s chaining together complex steps or cleaning up after itself when something goes wrong, GPT-5 holds its own like never before.

4. Multimodal mode unlocked.

You can toss it an image now. Soon? Audio and video. Big implications for accessibility, media, and creative work.

5. It hallucinates less.

That’s AI-speak for “makes up fewer facts.” GPT-5 is more grounded and transparent when it doesn’t know something.

6. Coding? GPT-5 eats it for breakfast.

It’s OpenAI’s best dev yet, writing, fixing, optimizing, and building full projects. Fast.

7. It’s getting agentic.

It can now act more independently, planning steps, using tools on its own, and tackling bigger workflows. Still early days, but you can see where this is headed.

8. Personal vibes are in.

You can tweak its personality, control how verbose it gets, and even connect it to things like Google Calendar. It’s starting to feel more like your assistant.

9. It comes in all sizes.

There are lighter versions (mini, nano) tuned for different use cases and devices.

10. You can try it now.

Even the free tier gets access. Devs also have new API endpoints to play with.

Bottom line? GPT-5 isn’t just another model. It’s a step closer to AI that actually gets you and does stuff for you.

But, that’s not all from OpenAI

Remember when OpenAI’s GPT-2 came out and the whole open-source community lit up with possibilities? Fast forward six years and they’ve finally cracked the vault again. This week, OpenAI dropped two open-weight language models: gpt-oss-120b and gpt-oss-20b and it’s the biggest nod to transparency we’ve seen from them since GPT-2. Both are available under the Apache 2.0 license, with downloadable weights ready for you to plug in, tweak, or run locally. No permission slips needed.

The 120b model flexes 117 billion parameters and runs reasoning tasks nearly on par with OpenAI’s o4-mini and it can run on a single 80 GB GPU. The smaller 20b variant is even more practical: it runs on machines with just 16 GB of memory. Think: edge deployments, laptops, or your weekend project that spiraled into something way more serious.

Under the hood, both use a Mixture-of-Experts (MoE) architecture for top-tier performance and efficiency. They support advanced capabilities like chain-of-thought reasoning, long context windows (up to 128k tokens), and few-shot function calling — serious tools for serious builders.

Now, this isn’t full open source. The architecture, training data, and routing logic are still under lock and key. But in the race against Meta, Mistral, and DeepSeek, it’s OpenAI saying, “Fine, let’s play open(ish).”

You can access the models on Hugging Face, Databricks, and AWS Bedrock. They’re not going to outshine GPT-4o or the rumored GPT-5, but if you’re working on reasoning-heavy tasks, they punch way above their weight.

Oh, and while all this was happening, OpenAI casually raised $8.3B to build AI agents that think like humans and plan like machines. Just another August week in AI land.

IN PARTNERSHIP WITH LEAH FROM ONEPROFILE
Still wasting time & $$$ to hire a dev?

Hiring great developers is a full-time job.

One most teams don’t have time for.

Between sifting through low-signal applications, coordinating interviews, and chasing no-shows, it’s easy to burn weeks (and serious budget) before you even make an offer.

That’s where Leah steps in ✊ a superhuman AI hiring agent built to find, screen, and hire world-class tech talent for you.

Leah sources globally, runs assessments, manages interviews, and surfaces only the top matches ready to deploy. Add her to Slack, and you’ll cut your hiring costs by up to 80% instantly ✂️

We’re opening up partnership opportunities for teams who want to start hiring right away with Leah on their side. If your team’s looking to scale smarter, not harder 👇

SUBSTRATES
What else is on the radar?

AI

• xAI launches Imagine, a new feature for SuperGrok and Premium+ users that turns text and images into fast-rendered videos with native audio, including a “spicy mode” for limited NSFW content.

• Character.AI launches Community Feed, an AI-powered social feed with users’ AI-generated videos, interactions, chat snippets, so others can remix or extend them with a tap.

• Perplexity teams up with OpenTable, allowing users to find and reserve tables through its platform.

• OpenAI rolls out new safety features to support users who may be in distress.

• Google announces the Kaggle Game Arena, a new AI benchmark where models compete in games like chess using text-based moves to test real-time reasoning, strategy, and adaptability.

• Google DeepMind introduces Genie 3, a new world model it sees as a key step toward achieving artificial general intelligence.

• ElevenLabs debuts an AI music generator that produces commercially-usable tracks, expanding its suite of audio-generation tools.

• Cloudflare alleges that AI startup Perplexity scrapes content from websites that explicitly forbid it while trying to hide its tracks.

WEB3

• Coinbase missed analyst expectations with $1.5B in Q2 revenue and a 39% drop in transaction income, sending shares down 7% despite crypto highs, as it shifts focus to staking, Base, and institutional services.

• Base Layer-2 surpasses Solana in daily token launches, with the shift beginning on July 24.

• The SEC clarified that liquid staking isn’t considered a securities activity, paving the way for platforms like Lido and Jito to operate without registration and sparking a surge in staking token prices and renewed ETF interest from issuers like BlackRock.

• Coinbase’s developer platform launches Embedded Wallets. This solution allows app developers to integrate self-custodial wallets into dApps with minimal coding.

FROM THE REPOSITORY
Check this out!

• The whole list of Meta’s Superintelligence team.

• Which public companies hold SOL the most?

• 10 thoughts on AI from Balaji Srinivasan.

Use now or save for later:

• Streamline and optimize your prompts with PromptGround.

• Keystone is the first AI QA engineer to join your team.

• Turn any website into an API with Parse.

*Shipping amazing products or services in AI or web3? Just hit reply and share your story with our 1k+ community—totally free!

Visual of the week: The evolution of stablecoins

SCOOPS OF THE WEEK