Claude Fable 5.2 solves physics problems and outperforms on security tasks

GPT-Live real-time voice enables viable AI tutoring at $3–5 per 30-minute session

Real-time voice capabilities are enabling immersive AI tutoring models competitive with human tutors: Mural's Norwegian teaching model uses GPT-Live to deliver sessions at $3–5 per 30 minutes, establishing a cost structure that undercuts traditional tutoring while signaling a new viable use case for voice AI infrastructure.

NYSE selects Avalanche for tokenized stock settlement, but critical chain design details remain undisclosed

NYSE announced Avalanche will run the core ledger for its planned 24/7 tokenized-stock venue (still pending regulatory approval), with trades matching on Pillar and settling on Avalanche. AVAX rallied ~7.5% on the news, but the announcement doesn't specify whether settlement will use Avalanche's public C-Chain or a private institutional chain—a critical distinction that determines whether AVAX holders capture meaningful transaction fees from this business.

Hollywood studios banning employee trading on prediction markets to protect insider information

Studios are restricting staff from trading on prediction markets after noticing massive disparities in betting volume (e.g., $2.3M on next James Bond vs. $1.0M on Apple CEO), revealing how prediction markets efficiently monetize insider knowledge and redistribute corporate intellectual property through unscrupulous traders.

Most stock issuers blocking OKX-backed tokenized stock venue while NYSE enables company-issued tokens

Of the 63 stocks OKXICE (ICE's 50-50 venture with OKX) initially proposed, issuers have now blocked 55 through SEC exemptions, whereas NYSE's competing venue lets companies issue tokens themselves through transfer agents like Broadridge and Computershare. This signals a structural preference for issuer control over third-party tokenization models.

Claude Fable 5.2 and Claude Mythos 5.2: Frontier Models with Improved Coding, Reasoning, and Safety

Anthropic published a product release for Claude Fable 5.2 and Claude Mythos 5.2, frontier models designed for ambitious technical work. Main takeaways:

• Fable 5.2 and Mythos 5.2 generate outputs 30% faster than prior versions at lower pricing ($8/$40 per million tokens vs. higher prior rates), with optional fast mode at 2.5x speedup
• On FrontierCode 2.0 agentic coding, Fable 5.2 at High effort matches Opus 5.5 at Max effort for roughly half the cost per task; at Xhigh effort scores highest of any released model
• Mythos 5.2 outperforms competitors on cybersecurity tasks: 89.6% on CVE Rediscovery (10.2% above Opus 5.5, 14.4% above Mythos 5.1); one security company found 1-in-6 bugs caught by Mythos were missed by all other frontier models tested
• Fable 5.2 solved two open problems in theoretical physics (black hole phase transitions), formalized Fermat's Last Theorem in 30 hours using 1,000 agents (vs. 11 days for prior approach), and designed a working prototype OS from scratch coordinating 1,000+ subagents
• Over 90% of code changes merged into Anthropic's core codebase now written by Claude; Fable 5.2 migrated critical production service from Python to Rust with 26x faster startup time

Madness and mayhem: A note on Jensen, AI spending, data centers & Anthropic

Exponential View published an essay analyzing the recent AI sector turbulence, Jensen Huang's security breach, and structural shifts in AI infrastructure economics. Main takeaways:

• AI revenue reached $188B in twelve months to October and runs at $345B annualized (4.2x year-over-year); Anthropic's slowdown to 45% quarterly growth reflects market normalization, not stall
• Democratic policy will increase data center permitting and construction costs, creating flight-to-quality favoring hyperscalers over neoclouds; 2027 capacity largely locked in and won't constrain revenue delivery
• Nvidia's China-by-proxy business (estimated 1M H100-equivalent units annually through Asian data centers) faces pressure; open-weight model strategy conflation with Chinese market access will reverse, favoring closed-model labs
• Anthropic IPO targeting $1.25-1.75T valuation before Thanksgiving would stabilize sector by shifting financing risk from Amazon/Google/Nvidia/SpaceX to public markets

Optimal virulence and the Android lockdown: the stalemate on general computation

dynomight published an original essay analyzing Google's planned Android app store lockdown through the lens of optimal virulence theory and competitive dynamics. Main takeaways:

• Google's lockdown addresses a real security problem (malware, social engineering) but uses self-serving centralized gatekeeping rather than alternative approaches like multi-entity certification or on-device scanning that would preserve user freedom.
• Current Android openness creates a strategic equilibrium: Google maintains near-total control while appearing permissive, preventing viable Linux phone alternatives from emerging and eliminating competitive pressure.
• If Google escalates to full lockdown, 300-500 million users currently sideloading apps outside Google Play could trigger a virtuous cycle toward de-Googled Android variants, creating existential risk for Google's ecosystem lock-in and forcing a choice between tolerating alternatives or aggressive suppression.

Meritech Software Pulse | 06-November-2026

Meritech Capital published a comprehensive weekly market analysis of the $3 trillion public software industry as of November 6, 2026. Main takeaways:

• AI-tailwind software companies up 142% since January 2024 (down from 199% two weeks prior) after post-midterm selloff; Palantir, CrowdStrike, Palo Alto Networks each fell 13-14% on datacenter/AI scrutiny concerns.
• Top 10 median ARR multiple compressed from 19.5x to 16.0x in two weeks; overall median fell from 5.0x to 4.5x; only 12 public software companies trade above 10x NTM revenue, 64% trade below 5x.
• Market regression analysis shows growth is 5.3x more strongly correlated with valuation multiple than FCF margin; companies with identical Rule of 40 but higher growth trade at 7.4x vs. 4.0x for lower-growth, higher-margin peers.
• Median net dollar retention stabilized at 107-109% range; median implied ARR per FTE rising as companies optimize for AI-driven productivity; payback periods ticked back up in H1 2026 after 2025 decline.