Topic
AI models distributed with public weights or permissive access, together with the platforms and communities that host and package them.
Tether open-sourced TranscribePsy, a 254M-parameter speech model it says transcribes 10 minutes of audio in 1.65 seconds on an iPhone 17 Pro Max. The accuracy and speed figures are Tether's own, published with evaluation configurations so outsiders can rerun them.
Reality
Broadcom CEO Hock Tan predicted the shortage of some data center chips will ease by 2028 or 2029 and could linger into 2030. Intel's Lip-Bu Tan spoke at the same Singapore event; the report does not include a date for Intel's supply to improve.
Reality
Cactus released Whistle, an open-source speech recognizer for seven languages that runs in 16.9 MB on a CPU with no GPU. Its speed lead over Whisper base was measured on one Apple chip, so teams building for wearables and microcontrollers need their own numbers.
Reality
New Constructs values Anthropic at $150 billion, less than a tenth of the roughly $2 trillion market cap discussed for its Nasdaq listing. The bear case rests on reported losses, and the price will depend on whether Anthropic's own revenue claims hold up in the prospectus.
Reality
China's STAR 50 has fallen about 30% since end-June, handing back roughly 70% of a nearly 75% three-month rally. Chip indices in Korea, Taiwan and the US fell with it, but the record fits cheaper open-source models and an unwinding rally as well as doubt about AI spending.
Reality
Featherless open-sourced Simple Jev, a library that has open models pick fixed labels or yes/no answers, with hosting from $0.03 per million input tokens. CEO Eugene Cheah pitches it for classification work teams now pay frontier-model prices to run.
Reality
OpenAI has cut input prices on its Luna models from $1.00 to $0.10 per million tokens since July 30, over two rounds of reductions. Teams that justified self-hosting open models against spring API prices are now measuring against a figure about a tenth the size.
Reality
The Lunar Foundation Model is trained by hiding parts of co-registered map tiles and making the network fill them back in, and its most interesting design choice is that it is told where the sun was rather than left to work it out from the shadows.
Perspective Coverage
3 publishersReality
Gartner puts this year's AI spending at $2.59 trillion, but the software and services layer the labs actually compete for is barely $1 trillion of it, which is what the compressed release calendar is defending.
Reality