Why this is in the vault
Wissner-Gross's sharpest framing yet of the efficiency-over-capability axis: the Singularity is arriving as a price collapse, and this issue logs the week it became undeniable across every major lab.
The core argument
The thesis is in the opening line: "The Singularity is proving as much a price implosion as an intelligence explosion." The issue marshals evidence across four weeks of AI launches to show that dollars-per-task has displaced benchmark supremacy as the operative competitive axis.
OpenAI GPT-5.6 (Sol/Terra/Luna). Three-tier launch — Sol (flagship), Terra (mid), Luna (price leader). Luna delivers GPT-5.5-level knowledge work at 10% of the cost. Sol runs at 750 tokens/second in Blender, clones Excel in six days, and posts ARC-AGI-2 at 92.5% — at one-tenth a three-month-old model's cost. Epoch suspects Sol is 5.5-sized, meaning gains are pure algorithmic efficiency. Sol trails Fable on SWE-Bench Pro; OpenAI responded by auditing that benchmark, finding 30% of entries broken, and retracting its endorsement — the benchmark Claude dominates.
Auto-posttraining. Sol autonomously post-trained Luna — a task that was a senior human team's job weeks ago. PostTrainBench: Sol at 50.3%, Terra at 51.5%. Noam Brown prefers 5.6 to a human intern. A developer vibe-coded a language model from his own iMessages.
Meta Muse Spark 1.1. Meta's first paid model, priced at roughly one-quarter of rivals. Leads on agentic tool-use; takes legal-agent SOTA. Zuckerberg declared war on "very extreme" lab margins. Analysts position Meta as the only hyperscaler world-class across data, talent, and compute simultaneously — projected to out-compute OpenAI and Anthropic combined by December. The frontier race went from three labs back to five in a single week; the efficiency frontier now has three seats: Fable, Sol, and Grok.
Anthropic's Veblen counter-move. Rather than racing to the bottom, Anthropic repositioned Fable on premium credits — roughly $10 in per million tokens, $50 out. Fable set a CIFAR-10 speedrun record rivals couldn't match. Elon Musk recanted public skepticism of Anthropic and vowed not to cut its rented compute. Anthropic shipped Reflect (an AI-reliance audit dashboard) and added Ben Bernanke to its trust board.
Substrate hardening. Micron committed $250B in US fabrication spend. Meta's custom chip enters production in September. SK Hynix launched a US IPO at $149/share. IEA projects AI will begin bending oil demand curves — first decline since 2020.
Governance gap. The AI Futures Project floated Plan A: delay superintelligence to 2040 via mutually assured compute destruction (MACD). Wissner-Gross's verdict: "Good luck rationing a price implosion. Don't decel." Europe passed Chat Control 1.0 through a procedural back door, scanning messages through 2028. Fed enlisted Andreessen on AI's economic impact. China graduated thesis-free practical PhDs and landed its first reusable rocket.
Physical-world acceleration. 1X unveiled a 25-DOF hand. Russian military trucks wearing drone-blinding dazzle stripes. SpaceX unveiled Starmind — a million-satellite inference swarm. Starlink hit 10 Gbps symmetric. Blue Origin raising $10B at $130B valuation. LA and NYC build flying-taxi vertiports for the 2028 Olympics.
Mapping against Ray Data Co
Anthropic's Veblen repricing (Fable at ~$50 out per million tokens) directly changes the cost structure for RDCO's COO agent loop — which runs Fable as its primary substrate daily. This is not a macro signal; it is an operating cost event. The $10-in / $50-out asymmetry makes output-token discipline (shorter responses, tighter context windows, subagent compression) the single most actionable efficiency lever RDCO can pull right now.
Secondary but high-conviction: Micron's $250B US commitment is a Phase 2 demand-signal for the memory cycle thesis. It lands the week after 2026-07-07-dram-hbm-phase2-phase3-early-signals.md logged zero triggers — this is the first named-dollar anchor worth tracking against the Phase 2→3 tripwire criteria.
Meta's margin-war framing (Zuckerberg vs. "very extreme" margins) is worth watching as a downstream pressure on Anthropic's Veblen strategy — if Meta captures enough enterprise share, premium positioning becomes harder to hold.
Related
- [[2026-07-07-dram-hbm-phase2-phase3-early-signals]] — Micron's $250B and the SK Hynix US IPO are the first named-dollar Phase 2 signals since this note was filed; update the tripwire log
- [[2026-06-16-innermost-loop-frontier-ai-token-price-index]] — OTPI tracks transacted token prices; this issue's price-implosion thesis is the qualitative counterpart to that quantitative index
- [[2026-07-08-every-efficiencymaxxing]] — parallel argument from the same week: the subsidized-experiment era is ending and ROI accountability is replacing tokenmaxxing
- [[2026-06-28-innermost-loop-singularity-horizon]] — prior Innermost Loop issue that first established the three-lab efficiency frontier framing this issue updates