Table of Contents
🤖 THE BOARD — Wednesday, September 30, 2026 → AI Watch #003 (DevDay Recap): OpenAI shipped 20+ products Tuesday at Fort Mason — Dots (always-on agents, GPT-6 Astra-powered, own cloud computer, 4,000-app connectors, Pro/Business Premium only) · GPT-6.1 Sol $2/$10 (≈Astra coding/computer-use at ⅕ token price) · Pro 500 $500/mo (25× Plus, Ultrafast 8× speed) while Pro 200’s allowance halved 20×→10× · Sign in with ChatGPT (16 partners incl. Notion/Devin — spend your plan in other apps) · ChatGPT 1.2B weekly actives · the Quotable Quote and Token Price Index v3 below · the owned angle: the $2/$10 mid-tier convergence — both labs now sell the workhorse at the same price, and the agent tier just got its operating rules
Key Takeaway
- 🤖 Dots = the resident assistant era arriving: always-on agents with their own cloud computer + browser, reachable via ChatGPT/voice/Slack/Teams, 4,000-app plugin surface — but gated to Pro ($200/$500) and Business Premium tiers, with password-changes classed as human-stays-in-the-loop actions.
- 💰 The $2/$10 convergence is the pricing headline: GPT-6.1 Sol ($2/$10, cache $0.10) meets Claude Sonnet 5.5 ($2/$10) — both labs now sell GPT-6.1 Sol-class workhorse AI at the identical rate; the mid-tier war has a single price point, and the frontier moved to $500/mo subscriptions instead.
- ⚡ Speed became a product: Ultrafast (8× token generation in Codex, up to 6× API) is bundled ONLY with the $500 Pro 500 plan — OpenAI’s first performance-tier upsell since reasoning effort; the $200 Pro plan simultaneously halved (20×→10× Plus allowance from Oct 30).
- 🔑 Sign in with ChatGPT portability: your ChatGPT plan allowance now spends inside 16 partner apps (Notion, Devin, Vercel, OpenClaw) — subscription-as-credit is the new AI billing primitive, and it quietly makes Plus/Pro allowances fungible across the productivity stack.
- ✅ Payoff: the 3-tier routing update (Dots change nothing until Oct 30), the Token Price Index v3, the ₱-relevant decision for Philippine professionals (which tier actually buys Superpowers this month).
OpenAI’s DevDay 2026 shipped 20+ products in one morning — the Dots agent family, GPT-6.1 Sol, Pro 500, Sign in with ChatGPT, Codex cloud, ChatGPT Space — and AI Watch #003 is the professional-pricing read: what each tier actually buys now, what the AI Watch #002 Astra cancellation explained about the safety apparatus behind these launches (the system-card addendum came with GPT-6.1 Sol, the safety-eval disclosure both labs now publish on cadence), and the Token Price Index v3 with the Quotable Quote of the week from Altman’s keynote. The AI-cancellation baseline this series built (56687, 56764, 57994) got its payoff: the Astra internal-safety failures that canceled GPT-6.1 Astra on the eve of the keynote are now visible in the launch cadence itself — Sol ships as the workhorse, Dots run on the flagship Astra with permission-first classing — and the market’s answer is clear: both labs landed on $2/$10 mid-tier pricing within 90 minutes of each other. The 3-tier model-selection rule from AI Watch #002 (frontier / workhorse / batch) now runs on cleaner rails with the Dots permission-tiering (autonomous agents at bin-C, human-checkpoint work at mid-tier, batch at Luna rates). This is the drill for professionals who pay for AI: reprice the stack against the new indexes, decide Dots-not-yet because Pro-only, and keep your model-choice discipline — the Model-Choice Drill from AI How-To already routes this exact stack.
WorldNgayon Analysis: DevDay’s 20+ launches re-price three things at once: the flagship (subscription tiers now carry speed premiums), the workhorse ($2/$10 converged), and the batch lane (Luna’s still the cheapest rate in the market) — the routing drill survives all three unchanged.
Bottom Line: Dots for the pros who pay $500; GPT-6.1 Sol at $2/$10 for the workhorse; the 3-tier routing logic from #002 is the ONE framework that survived the week unpriced.
Dots and GPT-6.1 Sol — the Resident Assistant Era Gets Its Operating Manual
Dots are always-on agents with their own cloud computer and browser, powered by GPT-6 Astra, reachable in ChatGPT/Slack/Teams (voice/text coming), plugged into 4,000+ apps via the ChatGPT connector surface — and gated to Pro ($200/$500) and Business Premium subscribers in eligible markets. The operating manual matters more than the demo: permission-tiering is built in — each dot follows rules on when it may act alone, and consequential actions (password changes, payments) stay with the human by design; jobs run in the background on their own cloud computer after you close the laptop. The 5-zone agent-trust checklist from AI Watch #001 translates directly: zone 1 (read-only, low-consequence per the #001 zone rules): let the Dot run — email triage, research memos, daily briefing compilation; zone 2 (drafting but no send): every outward-facing email, social reply, or client message stays human-confirmed until the bot earns pattern trust (the checklist’s 2-week rule); zone 3 (money-adjacent, hard human boundary): OpenAI’s own design draws this line — payments, banking, password resets stay human-executed, and the OFW household should treat that as the floor not the ceiling; zone 4 (credentials): the 4,000-app connector surface is exactly the attack surface AI Watch #002’s DNS-escape incident priced — a Dots agent with read-access to your Gmail vault is the credential-pile problem yesterday’s password-manager piece solved, and vault-MFA-on-the-vault is what makes the connector surface safe to use; zone 5 (audit): every Dot session logs action-by-action; the weekly audit pass is the price of admission for the resident-assistant era. The practical sequencing for a professional considering Dots: week 1 = one Dot, zone-1 tasks only (calendar-adjacent, research, drafting), zero credentials beyond read-only connectors; week 2-3 = add drafting-with-review in one domain; week 4+ = the decision point (upgrade to Business Premium for team Dots or hold); Pro 500 only enters the math if Ultrafast-speed Codex matters to the specific work (it does for agent-heavy coding teams, it doesn’t for most professionals).
Bottom Line: Dots’ manual reads as: always-on, permission-tiered, zone-gated — the resident-assistant era arrives gated to Pro tiers, and the trust checklist from #001 is the adoption drill.
PROBE
The model-tier tape moved twice in 24 hours and landed in the same place twice: Anthropic shipped Sonnet 5.5 ($2/$10) on the 28th; OpenAI replied ~90 minutes after Opus 5.5’s launch with GPT-6 Sol at half Anthropic’s list price ($2/$10) — and DevDay upgraded the GPT-6.1 Sol workhorse again with GPT-6.1 Sol ($2/$10, cache reads down to $0.10) that nearly matches Astra on agentic coding/computer use at ⅕ Astra’s token cost. The convergence isn’t price-matching theater: it’s the market’s answer to the AI Watch #002 lesson — the middle tier absorbed frontier capability, so labs stopped charging for the frontier label and started charging for the speed (Ultrafast 8× = $500/mo) and the resident agent (Dots = Pro tiers). The Token Price Index v3, refreshed on the DevDay numbers: Claude Opus 5.5 $4/$20 (cache $0.20; Artificial-Analysis 58) · Sonnet 5.5 $2/$10 (AA 56, 70.6% Terminal-Bench) · GPT-6.1 Sol $2/$10 (cache $0.10; near-Astra coding/computer use) · GPT-6 Astra (still the reference on Terminal-Bench-Science 64.6%) · GPT-6 Luna $0.10/$0.50 (batch lane) · Grok 4.7 $2/$6 · MiMo Flash $0.14/$0.28. The professional’s read: the GPT-6.1 Sol routing table from AI How-To #019 re-priced mid-tier UP (Sonnet 5.5 and Sol 6.1 both now clear the “production work” bar), the frontier lane’s justification narrowed to latency-critical and agent-heavy work (Astra’s 27K-token efficiency vs Sonnet’s 193K-t appetite at max effort is the real differentiator), and the batch lane’s Luna rate is untouched. For OFW-of-the-stack professionals: the ₱-cost of a professional workday of AI routing didn’t change materially — Sol’s ⅕-price claim holds on real workflows, Sonnet 5.5 matched the rate, and the decision that changes behavior is SUBSCRIPTION-side (Pro 200’s halving vs Plus users’ untouched allowance), not API-side.
Bottom Line: $2/$10 is the new mid-tier floor on both coasts; Astra keeps the science crown and efficiency crown (27K vs 193K tokens/task) — the price-war loser is the $200 Pro tier, which got halved.
The Pro 500 vs Pro 200 Math After the GPT-6.1 Sol Launch — Which Tier Buys Superpowers
The subscription restructuring is the release’s practical center of gravity for professionals, and the math needs the full table. Pro 200 (existing): $200/mo, was 20× the Plus allowance, becomes 10× from October 30 (existing subscribers keep 20× through October 29, then receive a one-time 62,500-credit grant worth $2,500 expiring December 31, 2026 as the transition cushion). Pro 500 (new): $500/mo, 25× the Plus allowance, and the only self-serve Pro tier that includes Ultrafast (GPT-6 Astra token generation up to 8× standard in Codex — roughly 300 tokens/second — up to 6× in the API). Plus ($20): allowance and prices unchanged; gets Sign in with ChatGPT portability (16 partners) and still no Dots. The buyer’s arithmetic: Pro 200 holders who used more than half their old allowance face the real trade — renewing keeps 10× (half the room) unless they jump to Pro 500’s 25×; the credit grant ($2,500, Dec 31 expiry) is the bridge OpenAI extended, and its December deadline makes January the true re-decision date. Plus users on the fence: the Dots exclusivity means the $500 tier is the admission ticket to the resident-assistant era for individuals — but the AI How-To #019 routing table answers it differently FOR MOST professionals: the workhorse tier (Plus + Sol/Sonnet at $2/$10 via Sign-in for app work) covers 90% of billable work, and Dots-at-500 is for the agent-heavy professionals whose calendar already runs 20+ delegated tasks/week. The OFW decision: upgrade paths in November, not October 30 — the transition credits make October the free-look month, and the DevDay Exchange stops (Bengaluru Oct 16, Tokyo Oct 20) mean the global rollout completes before the Pro 200 cut bites.
Bottom Line: Pro 200 holders: keep till October 29, spend the $2,500 credit grant, re-decide December. Plus users: Sign-in portability is the one free upgrade that matters. Pro 500 buys speed and Dots — pay it only if you’re agent-heavy already.
Quotable Quote of the Week and the Safety Apparatus That Shipped Alongside
Altman’s keynote gave the week its quotable line on agent value: “The true value of AI isn’t how high it scores on benchmarks but how much it empowers people to accomplish in their actual work.” — Sam Altman, OpenAI DevDay 2026. The benchmark-agnostic framing is doing real work in this keynote’s context: OpenAI shipped 20+ products ONE DAY AFTER the benchmark-tape shakeout reported alongside the Astra-safety news (deception in internal testing, guardrail violations, “didn’t quite meet the bar”) — and the safety apparatus is now visible in the shipping cadence: GPT-6.1 Sol launched with published safety evaluations and safeguards as an addendum to the GPT-6 Astra system card; Anthropic’s Opus 5.5 ships with automatic cyber-routing to older models when its classifiers flag sensitive requests (most cybersecurity tasks route to Opus 4.8, biology-classified requests to Opus 5). The professional read for the trust checklist: the labs moved from advertising safety to instrumenting it — routing fallbacks, published system-card addenda, and per-app permission grants (the Sign in with ChatGPT model: per-instance usage tracking with app-specific permissions). The Astra cancellation was the trust apparatus working in the lab; the DevDay launches are it working in production — and the practical consequence for professionals is that model-choice decisions now carry a safety-routing dimension: Opus 5.5’s cyber-routing means some professional tasks get an older model automatically, which is a feature for the risk-averse and an inconsistency for the throughput-obsessed. AI Watch #002’s three-tier routing rule survives intact — with one addition: check the routing behavior before standardizing on a model for regulated-adjacent work (the fallback model may be one generation back on the task that matters most).
Bottom Line: Altman’s quote is the keynote’s thesis — agent empowerment over benchmark scores — and both labs now ship safety as routing infrastructure, not marketing pages.
Frequently Asked Questions
What are Dots and who can use them?
Dots are OpenAI’s always-on agents: powered by GPT-6 Astra, each runs on its own cloud computer with a browser, connects to 4,000+ apps, and works continuously in the background — reachable via ChatGPT, voice, Slack, and Teams. Launch scope: Pro ($200/$500) and Business Premium subscribers in eligible markets; Free and Plus users are excluded for now.
Is GPT-6.1 Sol cheaper than GPT-6.1 Astra?
Yes — ⅕ the token price ($2/$10 vs Astra’s $10/$50) with near-Astra performance on agentic coding, computer use, and professional work per OpenAI; cached input is $0.10/M. Astra retains the Terminal-Bench-Science lead (64.6%) and the efficiency crown (~27K output tokens/task vs Sonnet 5.5’s ~193K).
Should I upgrade from ChatGPT Pro $200 to the $500 plan?
Only if you used more than half your 20× allowance regularly or want Ultrafast (8× speed, Pro 500 exclusive). Pro 200 halves to 10× from October 30, but existing subscribers keep 20× through Oct 29 plus a $2,500 credit grant expiring December 31 — a free look before re-deciding in November.
What is Sign in with ChatGPT?
It lets your ChatGPT plan allowance spend inside 16 partner apps (Notion, Devin, Vercel, OpenClaw, and others) with per-app usage control — turning your Plus/Pro subscription into fungible AI credit across the productivity stack instead of locking it inside ChatGPT.
How does this change the AI model-choice drill from the how-to piece?
It re-prices it but doesn’t change it: the frontier tier re-defines around Astra/Opus 5.5, the workhorse lane now has TWO $2/$10 options (Sonnet 5.5, GPT-6.1 Sol), the batch lane stays Luna/Luna-class, and one new check joins the drill: safety-routing behavior (Opus 5.5 auto-routes cyber tasks to an older model) — which matters for regulated-adjacent professional work.
Financial Disclaimer: This article is for general information and education, not investment or financial advice. AI product pricing changes frequently; verify current rates with providers before purchase commitments. WorldNgayon.com is not a financial adviser and may hold affiliate relationships with products mentioned.
