If you're charging human prices, having AI do the bulk of the work behind the scenes, and raking in the free margin, now is the time to ready a REAL differentiator of your business (i.e. it would still matter even if everyone knew exactly how you do what you do). This story jumping to #2 on the Radar today comes from $3.3 trillion (with a T) being erased from semiconductor stocks, and an upcoming weigh file that cannot be recalled (these are important and permanent numbers, and they make your most informed ICPs feel like you're scamming them if you don't have a true differentiator; long term, plan for everyone being informed). Today's LIVE AI Threat Radar has the full report and #2 is DEFINITELY worth a once-over. Its action plan is worth taking seriously and speaks volumes about the importance of non-lottery ticket differentiation, and the rise in importance and demand for model routing. One-line take: The price floor under "good enough thinking" stopped being a vendor's decision this week — once K3's weights are public on July 27, nobody can raise it back, so anything you bill for output rather than judgment is now priced against free.

"You can use up to 50% of your plan limits on Fable 5 through July 7. After that, switch to usage credits to continue using it." With no alternative in sight, Fable consumption-based pricing at standard API rates is currently Input: $10 per million tokens Output: $50 per million tokens Prompt caching: write is $12.50 per million tokens & read is $1 per million tokens Make it count now if you can't afford the change once it hits, but overall this points to the start of the next level of the AI Matthew Effect If you want to continue to be the first people to benefit from this type of AI, make your businesses capable of funding it and gaining from it the most. Then keep that cycle going

Despite restarts and /model workaround attempts, nothing has changed on my end, but apparently: US Commerce lifted the June 12 export-control directive on June 30 Claude Fable 5 — the general frontier model agencies feared — was restored globally July 1 after the US lifted its export controls; Mythos 5 returned with it, behind a new jailbreak-blocking classifier. Anticipation became fact overnight. Hits: execution & build work. Today's LIVE AI Threat Radar has the full report. One-line take: The two-week gated-models reprieve is over — the general frontier model agencies feared is back in clients' hands, so "who actually does the work?" gets harder to answer, even on a quiet news day.

The Radar surfaced something scary/exciting this morning: Axios reports it "on track to return soon," pending Pentagon/NSA sign-off) Fable 5 — the general frontier model agencies feared — is "on track to return soon" pending Pentagon/NSA sign-off; the NSA Director testified a sibling model autonomously breached nearly all NSA classified systems in a red-team; Anthropic's run-rate hits ~$30B with a science event today. Today's LIVE AI Threat Radar has the full report. One-line take: the two-week reprieve where the scariest tools were gated is ending — Fable 5 is nearly back and Google's no-government-leash edge is widening, so the pressure on execution- and content-led firms is about to step up again.