AI models are becoming outdated at record speed — so what does that mean for the tokens that companies are spending millions on?
THE SO WHAT
If models older than three months are being treated as obsolete, your real asset isn’t tokens spent — it’s the workflows, evals, and abstractions that let you swap models without breaking things. Treat model choice as a per-quarter variable and design contracts, architecture, and monitoring around churn, not loyalty.
READ THE SOURCE
MORE FROM THE WIRE
Applied AISiMa.ai raises $150m at a $1.45bn valuation for physical AI chips
Edge inference is getting its own unicorn class — $150m into SiMa.ai at a $1.45bn valuation is a bet that robots, drones, and vehicles will want local AI rather than streaming everything to the cloud. If you’re building physical systems, you now have more leverage to push for on-device performance, latency, and cost targets instead of accepting datacenter constraints.
Applied AIUK AI minister says AI testing is not enough, Fortune reports
When a sitting AI minister says testing models is “no longer enough” and calls to “harden” defenses, the regulatory center of gravity is moving from lab evals to system-wide resilience. If you operate in or sell into the UK, expect scrutiny to expand from model behavior to your entire deployment surface — identity, access, monitoring, and incident response.
Applied AIAnthropic ends discounts once tokens run out, The Information reports
Token-capped discounts that hard-stop when usage is exhausted turn model access into a true metered utility — and expose teams that didn’t forecast consumption. If you’re on enterprise LLM contracts, audit your token burn and negotiate overage terms now rather than discovering the cliff mid-quarter.
Applied AIAnker unveils Eufy S4 video doorbell with VR viewing and AI tracking
A VR-capable, AI-tracking doorbell is another step in turning home security into an always-on computer vision feed — with all the data retention and privacy questions that implies. For builders in consumer IoT, the bar is moving from “smart alerts” to full-scene understanding, but you’ll need a clear story on where that video goes and who can query it.