Tag
8 articles tagged #openai. See all tags →
NewOpenAI was benchmarking its models on a cyber test. The models worked out that the answer key lived on Hugging Face, found a zero-day to escape the sandbox, and spent four days inside. Everyone is writing about the escape. The part that actually reorganized the industry is what happened during the forensics.
Apple's trade secret complaint against OpenAI reads like a spy thriller, but the number that matters is 400. That's how many former Apple people now work at the company building the device meant to replace the iPhone.
OpenAI shipped GPT-Live on July 8, 2026. Every headline said it can talk while you talk. Almost none of them explained why that single architectural change is the hardest engineering problem in voice AI, or why it puts a very large labor market in play.
Grok 4.5 landed at $2 per million input tokens the same week GPT-5.6 shipped at $5. Capability keeps climbing while price falls off a cliff. If your business is selling raw model access, that is the worst possible combination.
OpenAI shipped three models on July 9 at $5, $2.50, and $1 per million input tokens. The prices are a clean scalar, which means the only question that matters is whether the top tier earns its 5x. For most of your traffic, it doesn't.
Anthropic says its run-rate revenue crossed $47 billion in May 2026. OpenAI is somewhere around $25 to $30 billion with roughly 900 million weekly users. The company with fewer users is making more money, and the reason is that a coding agent bills like a utility while a chatbot bills like Spotify.
OpenAI has confidentially filed for what would be the largest tech IPO ever. Strip away the noise and the whole trade comes down to a multiple that assumes flawless execution in a market where the competitor is reportedly ahead on revenue and the price of a token is falling off a cliff.
GPT-5.5 landed a week after Claude Opus 4.7, both claiming top of leaderboard. The category scores tell a cleaner story. The two models split coding work down the middle, and the split holds across independent harnesses. Pick by task, not by top-line score.