Anthropic’s Mythos is Back, OpenAI Releases GPT 5.6, Apple’s Price Increases

Episode Summary

Executive Summary: The episode examines how U.S. government oversight is increasingly shaping frontier AI releases, with Anthropic’s Mythos/Fable and OpenAI’s GPT-5.6 rolled out selectively to trusted partners first. The hosts argue this limits broad access, encourages routing toward cheaper/open-source models, complicates IPO narratives, and may weaken the closed-model business model. They also critique Apple’s steep price hikes as opportunistic amid memory shortages, while honoring the late Om Malik as a foundational tech voice.

Main Topics: Government control over frontier AI releases (Priority: 5/5): Anthropic’s Mythos/Fable release and OpenAI’s limited GPT-5.6 preview are framed as evidence that the U.S. government is becoming an approval gatekeeper for frontier models, raising concerns about who gets access and when. OpenAI GPT-5.6 capabilities and branding (Priority: 4/5): The hosts discuss GPT-5.6’s reported performance, its restricted rollout, and the odd naming of its variants (Sol, Terra, Luna), joking that the branding evokes failed cryptocurrencies. Business model and IPO pressure on closed AI labs (Priority: 5/5): They argue restricted releases and model routing may undermine the “expensive frontier model” growth story that investors have been buying, especially as OpenAI reportedly delays its IPO and Anthropic faces similar pressures. Shift toward cheaper models, routing, and open source (Priority: 5/5): Customers are increasingly using smaller, cheaper models and routing workloads by task, which the hosts say erodes reliance on frontier models and could boost open source and non-U.S. alternatives. AI billing opacity and cost overruns (Priority: 4/5): A report about customers reducing AI bills and an audit of billing errors highlights how hard it is to track usage and verify charges, reinforcing the sense that enterprise AI pricing is becoming contentious. Apple’s price hikes amid memory shortages (Priority: 4/5): Apple raises Mac and iPad prices by 15%–30% and is accused of using supply constraints as cover for margin expansion, with the hosts debating whether this is a normal market move or greed. Tribute to Om Malik (Priority: 3/5): The episode closes with a remembrance of Om Malik, emphasizing his generosity, influence on tech blogging, and lasting impact on the industry.

Key Arguments: The U.S. government’s selective approval of frontier models is effectively becoming a de facto regulatory system for AI access. Limited release may reduce misuse and model distillation, but it also concentrates power in the hands of governments and labs. If customers can’t rely on immediate access to frontier models, they may accelerate adoption of open source and model routing. The traditional AI IPO story depends heavily on rapidly growing spending on the most expensive models; routing to cheaper models weakens that narrative. OpenAI and Anthropic may want to preserve scarcity and premium pricing by restricting access rather than mass-releasing their best models. Billing opacity in AI usage makes it easy for overcharges and missed inefficiencies to go unnoticed. Apple’s broad price increases on devices with minimal memory content suggest it is capturing extra margin, not just passing through input costs. Apple’s ecosystem dependence may let it raise prices without immediate demand destruction, but the move is still seen as customer-hostile.

Data Points: Mythos access: 100+ U.S. institutions - Anthropic’s Mythos 5 is being released to more than 100 U.S. institutions, including major companies and government agencies. GPT-5.6 preview: Limited preview for a small group of trusted partners - OpenAI says broader availability will follow after coordinated testing and government-shared partner access. GPT-5.6 efficiency vs Mythos: ~100,000 tokens vs ~300,000 tokens - In an Exploit Bench cybersecurity test, GPT-5.6 was described as roughly 2–3x more token-efficient than Mythos Preview. GPT-5.6 price: About half the cost of Fable - The hosts note GPT-5.6’s preview pricing is roughly half that of Anthropic’s Fable. Ensemble Health annual AI spend: Up to $100 million - A customer cited in reporting plans to spend up to $100 million on AI this year. Savings from switching models: Nearly $700,000 per year - Ensemble Health said moving to a cheaper OpenAI model for insurance appeal letters would save about $700k annually. Monthly appeal letters: 15,000 - Ensemble sends about 15,000 insurance appeal letters per month. Model cost difference: 1/20th as expensive - A lower-cost OpenAI model was said to cost one-twentieth as much as the advanced model used for the same task. Billing audit: $1.7 million in mistaken overcharges - Vaudit audited $34 million in bills from 60 companies between March and June and found errors. Billing volume audited: $34 million - Total audited AI-related bills in the Vaudit report. Companies audited: 60 companies - Vaudit’s billing audit sample size. Micron gross margin: 84.9% - Used as a comparison point in the Apple greed debate. Apple gross margin: 49.3% - Cited to argue Apple has room to absorb some memory cost increases. Mac/iPad price hikes: 15%–30% - Bloomberg reported Apple raised prices on Macs and iPads by roughly this range. Base MacBook Air price: $1,299 - The base MacBook Air price reportedly increased from $1,099 to $1,299. Base MacBook Pro price: $1,999 - The base MacBook Pro reportedly increased by $300 to $1,999. iPad Air price: $749 - The iPad Air reportedly increased by $150 to $749. iPad Pro price: $1,199 - The iPad Pro reportedly increased by $200 to $1,199. Premium iPhone price increase over 7 years: 65% - Mentioned as evidence of Apple’s long-term pricing power.

Pivotal Quotes: "I think, basically, the government has installed itself as the approval of Frontier models." — Alex: A summary of the episode’s central concern about U.S. oversight of frontier AI releases. "It's a massive deal. The fact that it completely destroys the entire story." — Ronjohn Roy: On how cheaper routing and restricted frontier access weaken the IPO/growth narrative for closed AI labs. "Greed." — Ronjohn Roy: His direct response to whether Apple’s device price hikes are justified by memory costs.

Implications: Frontier AI may shift from broad consumer products to gated, premium infrastructure, boosting open source and routing strategies while complicating IPOs. Apple’s pricing move signals more margin pressure—and more consumer backlash—across hardware.