Collaborate with us Welcome back! Musk dropped Grok 4.5, Opus-class coding at a fraction of the price, free for now. ChatGPT's new voice talks like a person instead of a walkie-talkie. Google Omni will restyle your videos for a fee, and Anthropic is about to post AI's first real profit, right as Musk aims at it.
In today's Generative AI Newsletter: Grok: Can a $2 model really do Opus-level work?
It’s already and Opus level model but costs way less. $2 in, $6 out per million tokens. Opus 4.8 charges $5 and $25. It's the first Grok built for coding and agents, trained with Cursor on trillions of tokens of real developer sessions. Musk doesn’t want to overhype it tho. He admitted on X that it’s not close to Fable yet, but said you don’t need a Fable level model for most tasks.
Grok's pitch is the bill. It runs at 80 tokens a second and finishes the same task on a quarter of the tokens Opus burns. Musk calls it "roughly comparable to Opus 4.7, but much faster," and says the follow-up model, reportedly 2 trillion parameters, lands next month. If this one is already at an Opus level, what happens with the next one?
We’re excited. The new Grok is free for now in Cursor and Grok Build. The EU waits till mid-July. The AI Boom has a bottleneck but the solution is right in front of you Every AI company fights the same battle.
The market never sees most of them. For years, we built the system behind the scenes and successfully ran go-to-market for NVIDIA, Oracle, Google and 300+ AI brands, powered by data from a 14M+ community competitors can't reach. This helped us 2.5x our revenue last year, and now you can own a part of this AI distribution infrastructure. You can start from $1,000 and earn up to 22% bonus shares until August 31, 2026. Existing investors stack an extra 10% on top.
After August 31 the top tier drops to 20% and $1,000 checks stop earning bonuses. ChatGPT can speak better than humans now Every voice assistant so far has been a walkie-talkie. You talk, you stop, it answers. GPT-Live, which took over ChatGPT Voice yesterday, is a phone call.
It listens while it talks, decides many times a second whether to jump in or let you think, says "mhmm" while you ramble and goes quiet when you ask. The great thing here that it’s not one model doing everything at once anymore. While it’s listening and talking to you, it gives work to GPT 5.5 in the background to do the research and the reasoning for better outcomes. Same trick powers live translation.
Paid plans get GPT-Live-1 as the default, free accounts get the mini. No video or screen sharing yet, and the Hindi demo came back with an American accent. 150 million people already talk to ChatGPT every week. OpenAI's voice lead walks around having 30-minute conversations with it. The turn-taking era ended yesterday.
Google Omni is on your phone but not for free Google Photos got a new button. Video Remix takes any clip in your Create tab and relights it, swaps the background or repaints it in watercolor. Your dark concert video, fixed. No prompts needed.
It's for Google AI Plus, Pro and Ultra subscribers only, live in 14 countries including the US, India and Brazil. The EU isn't on the list. This is the Magic Eraser playbook again. Ship one irresistible trick, make it the reason people subscribe.
Google sells AI plans one video at a time. Anthropic is about to bank AI's first $1 billion profit Every lab burns money and calls it strategy. SemiAnalysis says Anthropic stops this quarter, with over $1 billion in operating profit, the industry's first. The engine is Claude Code, now touching 7% of all GitHub commits per the report.
Margins went from negative 94% two years ago to the mid-60s, and a megawatt that earned $16 million nine months ago is on track to earn $60 million. Around 80% of revenue is usage-based API. Anthropic filed for its IPO on June 1, so these are the numbers Wall Street prices. They're one research firm's estimates, so bring salt. The report's big line: keep executing and Anthropic could be the first $6 trillion company.
Tool of the Day: Willow Willow made AI dictation free and unlimited. The Frontier Mini model is cloud-based with zero data retention, and Willow claims it beats Wispr Flow, OpenAI and Deepgram on speed and accuracy. You talk, it types, in any app. Try this yourself: Download Willow for Mac, Windows or iPhone at willowvoice.com and pick the free plan.
Hold the hotkey and dictate a Slack reply, then watch it handle punctuation and names. Free covers unlimited dictation plus 20 Scribe uses a week. Pro at $15 a month unlocks the faster Frontier Pro model. Who it's for: anyone who types all day and talks faster than they type.
Everything else you shouldn't miss GPT-5.6 opens to everyone today : Sol, Terra and Luna, the models we covered yesterday, go public today after the Commerce Department cleared the release early. Meta's next glasses would photograph your whole day : The FT reports Meta is testing glasses that snap photos every few seconds and record audio all day, with no plan to light the capture LED. One day earlier, Meta published an FAQ boasting the LED "has no off switch." The bank that refused OpenAI now lends it $520 million : Bloomberg reports BofA extended its first credit line to OpenAI after earlier calling it too risky, a courtship move ahead of a $1 trillion-plus IPO. Beijing may let its champions buy Nvidia again : The Information reports Chinese officials told Alibaba, ByteDance and DeepSeek they may soon buy H200s, under a quota below 200,000 chips, less than half what they asked for. Cognition's new model writes code at 1,000 tokens a second : SWE-1.7, built on a Kimi K2.7 base, lands within a few points of frontier models and ships in Devin today, served on Cerebras. Learn more about AI from the experts building it 📸 Follow us on Instagram for fast, visual AI updates in 30 seconds. 📺 Watch us on YouTube to hear insights directly from leading AI voices, builders, and innovators. 🐦 Follow us on X for breaking AI news and real-time industry updates. 🧠 Learn how to build your next AI application with practical resources and expert guidance.