Anthropic’s release of Claude Fable 5 (commercial) and Mythos 5 (restricted) as a “rug pull,” arguing that after June 22nd users pay per token — they cite $10 per million input and $50 per million output tokens, roughly double Claude Opus 4.8’s cost — and get downgraded to Opus 4.8 on high-risk queries, which Anthropic reportedly says is under 5% of sessions. They claim the buried “real story” is that Anthropic will start “KYC-ing” users based on what it deems acceptable queries, which they call a privacy nightmare deliberately tied to the company’s upcoming IPO and pressure to pump its valuation. They urge running local AI models to preserve the ability to research and share “unpopular” data — crime statistics, immigration numbers, recession indicators — and link it to government crackdowns, naming Carney’s Canada and laws against “socially destabilizing content,” while debating whether Grok is the last uncensored platform.
Top 5 Key Topics
The “rug pull” and two-tier model split: The hosts say Anthropic split its newest model into Fable 5 for the public and Mythos 5 for elites, and argue that doing anything flagged as high-risk (cybersecurity exploits, chemical weapon design) silently downgrades users to Opus 4.8. They characterize this as burying the lead behind benchmark hype about coding being “68% better.”
Token pricing economics: They cite $10 per million input tokens and $50 per million output tokens — double Opus 4.8 — and note a colon or semicolon counts as two tokens while a letter is one. They argue a casual user’s budget might last weeks while a heavy user doing large research packages could exhaust it in hours.
Mythos 5’s elite, unrestricted sandbox: They claim Mythos 5 is a highly vetted, government-coordinated tier open to roughly 200,000 defense organizations and critical-infrastructure operators, where safety restrictions allegedly don’t apply. Fable 5, by contrast, is accessible to anyone via cloud API, Amazon Bedrock, Google Cloud, and Pro/Max/Team tiers.
“Shotgun KYC” and the IPO motive: The hosts argue the central danger is Anthropic moving to identity-verify users based on their queries, which they tie directly to IPO investors who they say can’t risk a “20% down day” if a bad actor is linked to the platform. They predict the 5% downgrade figure will balloon over time.
Run local, censorship, and Grok: They argue the only real defense is running AI locally to research and share contested topics, framing this against government efforts in the Five Eyes nations and Carney’s Canada to control what people can search, say, and share. They debate whether Grok is the last “reasonable” uncensored option, with one host skeptical that it isn’t quietly censoring too.
