Now showing on 15 screens
Happy Tuesday, legends. Welcome back to another edition of The Frontier, our weekly newsletter covering the best new AI launches on Product Hunt.
WHAT'S HOT
๐บ Anthropic ran a model across 15 smart TV boxes and the security world is arguing about it. Researchers split Gemma's layers over 15 networked boxes and, in a separate test, had a larger model deploy a smaller one, though that one still leaned on a server. The two were never combined into a working worm. Georgia Tech's Michael Bailey called it fearmongering, since self-replicating malware and botnets already exist. (Scientific American)
๐๏ธ Anthropic is putting 5% of itself into a Public Benefit Fund. Subject to a stockholder vote, with another 2% added at the board's discretion every time the company's value rises 50% above a set baseline. It lands ahead of an IPO expected in November. (AI Roundup)
๐ธ OpenAI is racing to close a round before Anthropic lists. Reuters reports an unusually fast raise aimed at locking in commitments ahead of Anthropic's mid-November debut, at a valuation that could pass $1.4 trillion. (Reuters)
๐ค Sam Altman confirmed OpenAI is building a humanoid. "I think we'll have a cool demo in 2027," he told the Sources podcast, while saying the body matters less than the brain that runs it. There is a lab for it and they are hiring for motion control, full-body behavior, manufacturing and supply chain. No features, no details, no robot. (India TV)
โ๏ธ Beam is a 501B open model built for agents. 23B active per token, 52 layers, 256 experts with 8 routed, a million-token context, BF16, FP8 and NVFP4 checkpoints, and reasoning that survives across turns. Apache 2.0, with its own Rust harness.
The week in products
FROM THE FORUMS
๐งพ Your code is not the expensive part
Mertcan logged one Claude Code session: 123k input tokens, 74% of them the list of tools, sent again on every request.
๐ฅ Every rule has a scar behind it
Enio's: never edit requirements.txt by hand. Added after an unpinned dependency jumped a version on deploy and took production with it.
๐ฐ A budget or a guess
Your agent is halfway through, it has spent $0.40, and nobody knows what the rest costs. Kushagra asks which number you should be holding.
๐
When did you last say no
Aadil's point: people stop reviewing the moment checking the draft takes longer than writing the reply themselves.
๐ Two customers, same price, different bill
Shaofeng on the awkward math of agent businesses, where one subscriber's retries cost ten times another's.