# Editorial notes — 29 September 2026

This is a proposed article, not a published post or trading instruction executed on an account.

## Catalyst verified as attributed reporting

The user supplied https://x.com/bubbleboi/status/2105060104479637895 . Followed its quoted post through Chrome/Playwriter to https://x.com/SemiAnalysis_/status/2105055045959995719 . The original SemiAnalysis account's post, dated 29 September 2026 at 3:00 PM in the browser's displayed timezone, explicitly names GPT6.1 Sol Ultrafast and reports Nvidia GPUs at low batch size instead of Cerebras. The draft now directly cites that post. Verified that SemiAnalysis made the claim; did not independently inspect OpenAI infrastructure. Capacity, preference, timing and stopgap intent remain interpretations.

OpenAI's 29 September release explicitly says Astra Ultrafast is available today and Sol 6.1 Ultrafast is coming soon. The draft preserves that distinction. The August Cerebras-powered 5.6 Sol preview is acknowledged, so this article does not claim Cerebras never powers serious OpenAI launches. No claim about silence from either company is made from absence alone.

## Sources consulted

- https://x.com/SemiAnalysis_/status/2105055045959995719 — directly read in authenticated Chrome, original catalyst report.
- All seven articles listed at https://corbanu.com/posts/ were fetched as current style/context references. Main research basis: original Cerebras short, CXMT sizing, and the latest Apple/Oracle/SK hynix notes. Local source captures are retained in the original research workspace. The original financial model is explicitly historical, not refreshed consensus.
- https://openai.com/index/introducing-gpt-6-1-sol/ — model positioning and availability.
- https://openai.com/index/devday-2026-recap/ — Astra/Sol Ultrafast availability distinction.
- https://openai.com/index/previewing-ultrafast/ — August preview powered by Cerebras; up to 750 output tokens/sec, limited availability.
- https://openai.com/index/cerebras-partnership/ — 750 MW, phased through 2028.
- https://newsletter.semianalysis.com/p/openai-jalapeno-better-than-nvidia — background only, not evidence of the September 29 hardware report.

## Sizing and scope

3.5% of capital in short notional is the proposed model-portfolio target, twice the 1.75% published original. It is not a claim about present holdings, a new entry price or an executed adjustment. At unchanged initial notional, a 50% adverse price move costs 0.035 × 0.50 = 1.75% of capital before fees/funding. This deliberately adds a directional bet; it is not a refreshed beta-neutral sizing calculation. No current funding or entry quote was retrieved. The old $118 target is identified as the original model's base case, not a fresh mark-to-market valuation. Before adoption reconcile current position status, quote/funding and the original stop policy against the proposed change.

The requested argument and themes were supplied directly, so no separate theme-selection or Claude harness run was commissioned. Applied the existing Corbanu house profile and selective Humanizer editing directly. No skill was changed. The user subsequently approved a GitHub draft branch and a shareable Cloudflare preview; see README.md for delivery details.