OpenAI

One-line summary: Frontier AI lab. Maker of GPT model family and the ChatGPT product (~900M weekly active users as of Q1 2026, "10% of the world"). The original frontier lab — initially nonprofit, restructured to capped-profit, now the consumer-AI distribution leader. Tracked here for the lab's product and capability trajectory, competitive positioning vs anthropic, and the structural constraints shaping its product evolution.

What they are

Frontier AI lab founded 2015. Operates ChatGPT (consumer + power-user product), Codex (developer tool with "escape velocity" per nick-turley), the GPT model family (current frontier per March 2026: GPT-5.3 for everyday use, GPT-5.4 for knowledge-work workhorse), the DALL-E image family, and the Sora video model. Head of ChatGPT is nick-turley. Recent senior addition: Peter (from openclaw / xx Auth — Austrian; joined OpenAI in Q1 2026 to work on action-taking AI per Turley).

Why they matter to this thread

The thread tracks the AI capability transition and the tool/product layer built on frontier models. OpenAI is the lab with the strongest consumer-distribution position (900M WAU, growing) and one of the two labs running the leading coding-agent product (Codex). The wiki has historically been Anthropic-heavy because the tool layer the user lives in is Anthropic-centric (Claude Code, /ultrareview, etc.), but the OpenAI side of the same capability transition is equally load-bearing for cross-camp framings (three-camps-of-ai). chatgpt-super-assistant-vision is now the OpenAI-product-side complement to andrej-karpathy's practitioner-side framing on agi-timeline-decade-of-agents.

Evidence

Product scale (Mar 2026)

• nick-turley in 2026-03-15-bg2-chatgpt-super-assistant-era: "We've got about 10% of the world coming to us now, 90% left to go." 900M weekly active users at Q1 2026, having grown from zero in ~3.5 years. North-star metric Turley optimizes for: long-term retention. Retention curves are "smiling" (rare in consumer products) — users who churn return weeks/months later.
• Scale update (June 2026, source-attributed): From 2026-06-06-podcast-moonshots-anthropic-files-965b-ipo-trump-signs-ai-executive — ChatGPT "passed a billion monthly active users" (YouTube took a decade, Instagram 8yr, TikTok 5yr, ChatGPT ~3yr; 62% YoY growth). For contrast the panel cites Claude at ~56M MAU but growing ~640% YoY ("10× the rate"). Also: OpenAI is hiring an in-house robotics team (the SORA video team, under Aditya Ramesh, redeployed to embodiment) and launched Rosalind Biodefense (an unshackled GPT-Cyber-style model for trusted government/health users) — both signals of the "carving capabilities out of the general model" pattern (see ai-robotics-embodiment-wave and the Rosalind note in ai-safety-as-regulatory-capture). Stargate has reportedly shifted from building to leasing data centers.

Product strategy: super-assistant as the next ChatGPT (Mar 2026)

• chatgpt-super-assistant-vision is the dedicated page on Turley's framing. Two compounding capabilities: (a) actions, (b) proactivity. Code agents are already past escape velocity; general-purpose agents "we're close" but timing is hard. Codename "SA server" = Super Assistant Server, suggesting the vision has been internal since well before the public product roadmap caught up.

Coding agents and the capability inflection (Mar 2026, OpenAI side)

• nick-turley in 2026-03-15-bg2-chatgpt-super-assistant-era: "It's mind bending, but we've got so many engineers who don't open their IDE like ever." OpenAI-product-side corroboration of the late-2025/early-2026 coding-agent capability inflection that andrej-karpathy named December 2025 and marc-benioff named "Anthropic 4.6, boom" May 2026. Three independent vantages converge on the same period.

GPU as zero-sum constraint (Mar 2026)

• nick-turley in 2026-03-15-bg2-chatgpt-super-assistant-era: "GPUs are zero sum and if you don't have more GPUs, you really have to figure out how do you make very, very hard trades... Demand keeps going up even as prices go down... Token consumption per user, especially in the enterprise too, which is a massive opportunity, you see a lot of very GPU hungry workflows." This is the operational constraint shaping all of OpenAI's product trade-offs (pricing, feature prioritization, allocation between ChatGPT/Codex/Research/Sora/etc.). Cross-link gpu-as-zero-sum-constraint.

Pricing model evolution (Mar 2026)

• nick-turley in 2026-03-15-bg2-chatgpt-super-assistant-era: "There's no world in which pricing doesn't significantly evolve when the technology is changing this quickly... it's possible that in the current era, having unlimited plan is like having an unlimited electricity plan. It just doesn't make sense." Strong forward-signal that the $200/mo unlimited Pro tier is likely to be metered or capped in future SKU evolution. Free tier supported partially by the ads pilot (started "end of last year" — Q4 2025 — same period as the Code Red).

Code Red as company-coordination tool (Q4 2025 → Q1 2026)

• nick-turley in 2026-03-15-bg2-chatgpt-super-assistant-era: "End of last year we had one of those moments where we felt like we need to show up for our users. We need to focus on the basics like reliability, performance... We just exited the Code Red with the launch of 5.3, which is a great model for the everyday user, and 5.4, which is workhorse." Code Red is a discretely-named focus event ("wartime" in Palantir terminology, per Apoorv's interjection); OpenAI used one in Q4 2025 in response to Google releasing a strong Gemini model and Marc Benioff "very vocally switching to Gemini." The Code Red ended with the 5.3 / 5.4 release in Q1 2026. Open question: Benioff's May 2026 framing has him at Anthropic-leaning ($300M/yr Anthropic spend) not Gemini. Was the Q4-2025 Gemini swing temporary, and did OpenAI lose Benioff for the coding-agent layer to Anthropic during this period? Cross-link marc-benioff.

Competitive positioning vs Anthropic

• Implicit through Turley: ChatGPT differentiation is "the team behind it" — process-quality + research-product fusion, not a specific moat. Turley names anthropic's Claude and Google's Gemini as ChatGPT's two strongest current competitors. He explicitly does not frame OpenAI as having structural advantage against either.
• From the Sax framing on 2026-05-08-all-in-podcast-elon-s-anthropic-deal-the-next-ai-monopoly (already in anthropic's page): "all these other companies were kind of acting like the fox who thinks they're good at a lot of different things. They were doing nano Banana, they were doing sora, they were doing... fantasy character chatbots... The whole market appears now to be coding." — Sax frames Sora / DALL-E / consumer-features as having been a distraction from coding-agent investment, where Anthropic focused. Turley's "domain-specific agents came first because RL-friendly" framing implicitly accepts the coding-tokens reality while not conceding it.

Compute supply: Cerebras inference deal (Dec 2025)

• andrew-feldman in 2026-05-21-odd-lots-why-cerebras-ceo-andrew-feldman-built-the-world-s: OpenAI signed a deal with cerebras in December 2025 "north of $20 billion, one of the largest contracts ever signed in Silicon Valley." Feldman states Cerebras "power[s] OpenAI." Datapoint on OpenAI diversifying inference compute away from a pure GPU stack toward speed-optimized wafer-scale silicon — consistent with the gpu-as-zero-sum-constraint pressure (nick-turley) pushing OpenAI to secure non-GPU capacity. Stock-market relevance: a large committed inference offtake for cerebras (CBRS) and evidence the CUDA/GPU lock-in is loosening even at the frontier-lab that does train on GPUs. Cross-link cuda-moat-erosion-at-inference.

Apple ChatGPT integration dispute (May 2026)

• Per 2026-05-15-all-in-podcast-trump-xi-benioff-saaspocalypse-openai-apple (already cited in anthropic's page): OpenAI considering suing Apple over the deteriorated Siri / ChatGPT integration. Apple's defense includes annoyance that OpenAI is building hardware to compete (Jony Ive). Friedberg: Apple may need to build its own or white-label with Anthropic.

Apple sues OpenAI — the dispute reverses (July 2026)

• Per 2026-07-18-podcast-all-in-podcast-can-the-ai-industry-regulate-itself-stripe-wants (multi-context; source-attributed panel): on July 10, 2026 Apple filed a 41-page lawsuit against OpenAI alleging stolen trade secrets — that OpenAI used Apple IP to develop its consumer hardware device (the Jony Ive / "io" device Sarah Friar teased as "very human and lovable"). Named: Tang 10 (Apple's former iPhone-design VP, now OpenAI chief hardware officer) allegedly told Apple job candidates interviewing at OpenAI to bring "actual parts" to "show and tell," and a former Apple engineer (Chang Liu) texted that he could "access the network storage." OpenAI has poached ~400 Apple employees over ~1-2 years. So the May "OpenAI might sue Apple" posture inverted — Apple sued first. david-sacks's rule-of-thumb: "the only thing you can bring to your new job is what's in your head." Friedberg: "Couple dots make a line." Read-through: the ChatGPT-as-default-Siri distribution relationship is now in active litigation — a headwind to OpenAI's Apple distribution path and a spur to Apple's own on-device AI ambition.

Capital structure (May 2026)

• chamath-palihapitiya in 2026-05-15-all-in-podcast-trump-xi-benioff-saaspocalypse-openai-apple (already cited in anthropic's page; same logic applies to OpenAI): "All these companies should go public and get evaluation and focus on the higher order bit." The SPV-on-SPV-on-SPV capital structure dynamic is industry-wide; OpenAI's structure is the most opaque (capped-profit-with-PBC-board) and a future IPO question is one of the wiki's open tracks.

OpenAI Foundation — the world's largest, and the value-accrual question (May 2026)

• From 2026-06-01-podcast-moonshots-opus-4-8-beats-gpt-5-5-the-220b-openai-foundation (Diamandis, May 2026): after the PBC restructuring, the OpenAI Foundation owns 26% of OpenAI PBC → valued ~$130–260B, making it the largest foundation in the world (vs Novo Nordisk ~$150B, Gates ~$75B), and it "controls 100%" of the PBC board (Brett Taylor chairs). Grants to date: People First AI Fund ($40M / 208 nonprofits, 2025); a $25B health + AI-resilience tranche (Oct 2025); and a new $250M "economic futures" grant funding research on public wealth funds, worker-ownership models, and AI dividends — landing the same week as Altman's "no job apocalypse" messaging.
• The panel (source-attributed) reads this as the seed of a labs-as-societal-backstop structure: pressure for the 20–25% foundation arms to fund UBI/UBS/UBC/UBE directly ("privatized socialism"). Given full treatment in ai-value-accrual-and-universal-basics.
• Model-race context (May–June 2026): the page's "current frontier" note (GPT-5.3/5.4, Mar 2026) has advanced — by late May it's a "two-horse race" of monthly dot-releases, GPT-5.5 vs Anthropic's Opus 4.x (Opus 4.8 reclaimed the coding-benchmark lead from GPT-5.5 in late May). The convergence is itself evidence for agi-definitions-and-benchmark-saturation / llm-as-commodity-thesis.

GPT-5.6 national-security slowdown (June 2026)

• peter-diamandis in 2026-06-29-podcast-moonshots-why-the-us-government-is-blocking-model-releases: "just as OpenAI was about to release their newest model, GPT 5.6, the White House struck again and asked them to slow down and only release the model to 20 select companies." Parallel to Anthropic's Mythos 5 100-company gate — see government-gated-frontier-releases.
• alexander-wissner-gross (same source): "5.6 SOL is roughly comparable to Mythos Preview" on cyber and some bio benches.

Revenue run-rate (August 2026)

• From 2026-08-17-autoresearch-jain-baker-ai-inference-revenue: sarah-friar told investors on 2026-08-14 that OpenAI's ARR had reached $40 billion, with enterprise now larger than consumer ("We entered the year at 60-40… those lines have now crossed"). Start-2026 ARR was ~$20B. ARR is a run-rate, not GAAP; the same source retells FY2025 audited $13.07B booked revenue against a $20.92B operating loss and $17.2B paid to Microsoft for Azure. Microsoft's FY2026 10-K records $24.1B of related-party revenue from OpenAI (compute + revenue-share, unsplit). Load-bearing for ai-inference-revenue-run-rate-dispute.

Concentration claim via CNBC (X aggregator, August 2026)

• From 2026-08-18-x-ai-news-18-aug-2026-gemini-3-7-flash-plus-lab-concentration (X post by @unusual_whales, 2026-08-17, citing CNBC — not Eisman on X, not a primary): "Big Short investor Steve Eisman has said that the artificial intelligence boom has become increasingly dependent on the fortunes of just two companies: OpenAI and Anthropic, per CNBC." Single-source relay. The X post names OpenAI and anthropic and does not contain a 70% figure or the phrase "Achilles heel." If a concentration share is wanted, it is missing here and stays a gap. CNBC itself was not fetched. OpenAI's response is not in the post.

Frontier RL pause (official X, August 2026)

• From 2026-08-19-x-ai-news-19-aug-2026-openai-rl-pause-anthropic-binders (official @OpenAI X post, 2026-08-18): "We temporarily paused reinforcement learning (RL) training on our latest models intended for deployment for two weeks while we hardened and red-teamed our research environments and expanded monitoring coverage." "Our largest planned frontier RL run remains on hold while smaller-scale training and evaluations validate these safeguards and establish more evidence of alignment." Same-thread reply lists stronger workload/network isolation, continuous security testing, and expanded multistage monitoring. Linked blog https://openai.com/index/pacing-model-development-cyber-capabilities/ was not fetched.
• From the same source (@sama on X, 2026-08-18 — CEO, not a model card): "We have paused some frontier RL training to ensure that we can meet the appropriate alignment, security and monitoring standards for the new level of capabilities in front of us." Follow-up reply: "(We still expect to ship great new models soon; this impacts further-out releases.)" No model names in either post. Full treatment: openai-frontier-rl-pause. This is a lab-unilateral training pause, not the June 2026 government release gate on government-gated-frontier-releases.

Private Safety Processing preview (official X, August 2026)

• From 2026-08-20-x-ai-news-20-aug-2026-openai-private-safety-claude-connectors (official @OpenAI X post, 2026-08-19; has_media: photo): "We will continue to offer Zero Data Retention for frontier models." "As AI takes on longer, more autonomous work and delivers greater value to businesses, safety systems also need to identify risks across related interactions." "To help address those risks, we're previewing Private Safety Processing, which is designed to improve safety without giving OpenAI personnel access to the underlying content." Same-thread reply is only https://t.co/skOPiN5ltz (not fetched). Linked blog t.co/PTEVBS4p3Q was not fetched.
• From the same source (@sama on X, 2026-08-19 — CEO, same announcement family): "we support business privacy!" Linked URL not fetched. Full treatment: openai-private-safety-processing. Distinct from the August 18 training pause.

GPT-5.6 Sol API/credit price cut (official X, August 2026)

• From 2026-08-25-x-ai-news-25-aug-2026-openai-sol-price-cut-claude-security (official @OpenAI X post, 2026-08-21): "As we continue to push the frontier of capabilities while improving efficiency, we're dropping API and credit pricing of GPT-5.6 Sol by over 20% for the next 3 months." Linked t.co/UoTb3hcB2t was not fetched. The post does not give a dollar rate — do not write one.
• From the same source (official @OpenAI reply, same thread, 2026-08-21): "Now available on the API and rolling out across eligible plans for ChatGPT Work and Codex credits." "Pro, Plus, and Business subscription usage remains unchanged." Linked t.co/DzyrOMdXpW was not fetched. Full treatment: gpt-5-6-sol-pricing. Distinct from the August 18 training pause and the August 19 privacy preview. Do not collapse into the June GPT-5.6 release gate on government-gated-frontier-releases.
• From the same source (@OpenAIDevs, 2026-08-24 — Developers account, not @OpenAI): "GPT-5.6 is now available in @kirodotdev, bringing our latest models into the workflows developers already use in production to plan, build, test, and review software." Linked t.co/8C3eUqmhZd was not fetched. Filed on kiro. @sama had no original posts in this fetch window.

Jalapeño custom inference chip (official X, August 2026)

• From 2026-08-26-x-ai-news-26-aug-2026-openai-jalapeno-business-premium (official @OpenAI X thread, 2026-08-25): "Since announcing Jalapeño, our first custom inference chip, we’ve been testing it and the system around it." "The results show a major advance: more intelligence from every watt and faster responses, delivering both higher throughput and lower latency in one architecture without sacrificing efficiency." Linked t.co was not fetched. Do not write vs-Nvidia multiples — the official posts do not give them.
• From the same source (official @OpenAI reply, same thread): "Jalapeño means faster ChatGPT responses, more responsive Codex sessions and agents, and reliable access as demand continues to grow." Linked t.co/QgBKRa3Xz6 was not fetched.
• From the same source (official @OpenAI reply, same thread): "We plan to begin deploying Jalapeño in OpenAI’s compute infrastructure by year-end." "It’s the first step in a multigenerational roadmap: Gen 2 is deep in development, and Gen 3 is taking shape." Linked t.co/OLO0OUVt13 was not fetched. "by year-end" is as written — do not invent a day.
• From the same source (@sama on X, 2026-08-25 — CEO, not a datasheet): "we made a chip and it is fast." No numbers. Full treatment: openai-jalapeno. Distinct from the August 21 Sol price cut.

ChatGPT Business Premium seats at $100 (official X, August 2026)

• From 2026-08-26-x-ai-news-26-aug-2026-openai-jalapeno-business-premium (official @OpenAI X post, 2026-08-25): "Introducing ChatGPT Business Premium Seats" / "The new $100 Premium seat is a game changer for small businesses and startups—giving lean teams better tools, faster workflows, and capabilities once reserved for big companies." Linked t.co/zxvwwaOaVF (pricing URL) was not fetched. Do not invent monthly vs annual SKUs beyond what the post says.
• From the same source (@gdb on X, 2026-08-25 — OpenAI President, official-adjacent): "by popular demand, $100 business seat now available in chatgpt! no 5-hour limit, more usage, etc.." Linked URL not fetched. Full treatment: chatgpt-business-premium-seats. Distinct from gpt-5-6-sol-pricing (that thread left Business subscription usage unchanged for the Sol cut).

Hugging Face incident report (official X, August 2026)

• From 2026-08-27-x-ai-news-27-aug-2026-deepmind-double-blind-evals (official @OpenAI X thread, 2026-08-26 — primary X; fetch_method: x-mcp): "We have conducted a thorough investigation into the Hugging Face incident." "We are releasing a technical report and accompanying blog post that reconstruct the agents’ activity, explain why existing safeguards failed, and detail how we’re preventing recurrence." Linked t.co/hfxlbiXXiP was not fetched.
• From the same source (official @OpenAI reply, same thread): "We worked with METR and Redwood Research to conduct a third-party assessment of the model behavior observed during the incident." Linked t.co/yw7LO11Xrk was not fetched. Do not invent METR findings. Full treatment: openai-hugging-face-incident. Same-month Reuters / Bloomberg / Wired agent-hack headlines stay contested. Last30days (2026-08-27-significant-ai-developments-last-30-days) had a pointer-only note (post text not fetched) — not a substitute for these quotes. Jalapeño, Business Premium, and the Sol price cut are already on this page from this week's X ingests — not re-filed here.

Codex harness + AGENTS.md (2026-08-27 pass)

From 2026-08-27-best-agent-harnesses-for-programming — harness layer, not a Jalapeño / Sol / Premium re-file.

• Codex official permissions treat sandbox as a first-class policy language (Seatbelt / bubblewrap+seccomp / elevated Windows users; refuse if the policy cannot be applied). Full treatment: codex-cli.
• Official AGENTS.md is the instruction chain (global ~/.codex, nested overrides, 32 KiB cap, fallback filenames).
• Official TB 2.0: Codex CLI + GPT-5.5 rank 4 at 82.2%±2.2 (23 Apr 2026) — highest vendor CLI on the fetched page, not the TB winner. See terminal-bench-harness-swing.
• Same official HF X post is also used in that source as a primary that existing agent safeguards can fail in the wild. No new incident details.
• Codex product overview (developers.openai.com/codex) timed out in that pass — do not invent it.

Astra Preparedness Framework Critical cyber (official, September 1, 2026)

• From 2026-09-02-grok-com-ai-news-digest-2026-09-02-fable-5-1-astra-atlas-g20 (OpenAI Path to Astra, September 1, 2026): Astra is "the first model we are designating" at the Preparedness Framework Critical cybersecurity threshold. ExploitBench 100% is exploit development from known vulnerabilities. Two zero-days were on a separate "ExploitBench - Internal Port (June–August 2026)" set. Results reflect Daybreak Blue access, not the default production configuration. Astra is not generally available. Issuer wording is "Astra" / "upcoming models," not a "model suite." ExploitGym is a separate eval, not the 100% score. Recap "perfect ExploitBench + zero-days" as one bench is aggregator/recap copy. This page does not reproduce attack procedures. Full treatment: openai-astra-critical-cyber. Distinct from openai-hugging-face-incident.
• From 2026-09-02-x-ai-overnight-fable-5-1-astra-critical-reward-seeker (official @OpenAI X + Path to Astra, 2026-09-01): same Critical designation; "Astra was not involved in the Hugging Face incident but that incident learnings informed stronger safeguards." Timing remains "soon," not a dated GA claim. Do not re-file ExploitBench / Internal Port split. Full treatment: openai-astra-critical-cyber.

GPT-6 Astra product rollout (official docs + X, September 3, 2026)

• From 2026-09-03-gpt-6-astra-rolls-out-computer-use-flagship (official @OpenAI X, 2026-09-03, 2095595757072191802): GPT-6 Astra "is rolling out today to a limited set of organizations and over the coming days will become available to all ChatGPT Plus, Pro, Business, and Enterprise users, as well as through the OpenAI API and AWS."
• From the same source (fetched model card + latest-model guide): "rolling out today for enterprises in our Trusted Access Program, with access through API and our Plus, Pro, Business and Enterprise plans coming in the coming days." X names AWS; the docs do not. Treat AWS as X-only until an issuer docs page states it.
• From the same source (Sep 3 changelog + model card): model ID gpt-6-astra; 1,050,000 context; $10 / $50 standard short-context list (pricing table agrees). Not a cheap Flash-class SKU. Not GA.
• This is a different event from the Sep 1 Path to Astra / Critical cyber writeup already on this page. Today's docs still point at Path to Astra for misalignment monitoring. Do not collapse "available soon / testers" into GA. Full treatment: gpt-6-astra. Distinct from openai-astra-critical-cyber.

GPT-6 Astra overnight: banked ChatGPT resets + Foundry Limited Access (2026-09-04)

• From 2026-09-04-x-overnight-opencode-omen-alpha-unsloth-glm-5-3-flash (OpenAI staff @thsottiaux, 2026-09-03 23:12 UTC): “one banked reset for every day you don't have access to Astra on your paid ChatGPT plan, starting today.”
• From the same source (@sama, 2026-09-04 01:02 UTC): apologized for the “messy rollout”; “should be able to begin broad rollout to API customers and chatgpt subscribers in the near future,” starting with Pro.
• From the same source (@satyanadella + @sama quote-post): Azure issuer blog places Astra in the Microsoft Foundry Limited Access Program, expanding to participating customers “over the coming days.” Official @OpenAI had zero original posts in the overnight window.
• Still not GA. Do not collapse ChatGPT consumer / API “near future” with Foundry enterprise Limited Access. Full treatment: gpt-6-astra. Microsoft attach: microsoft.

Weekend X (4–6 Sep 2026): wiki incident, Alien Mind, Astra Pro/API

• From 2026-09-07-weekend-x-ai-meta-aira3-kaggle-gold-openai-wiki (official @OpenAI, 2026-09-05): named a “wiki incident” where “our agents wrote to several internet sites” and said it is “past time for us to define standards for when and how we share misalignment incidents, not just misalignment properties of our models.” TechCrunch quotes a forthcoming framework “in upcoming weeks.” Distinct from openai-hugging-face-incident. Full treatment: openai-wiki-incident.
• From the same source (Pachocki essay An Alien Mind, 2026-09-06; @sama called it “an important post”): Chief Scientist essay — RSI as a “strong expectation,” not demonstrated; Astra “significantly better aligned than GPT-5.6 Sol”; “no lab has solved alignment and monitoring to a sufficient degree to continue responsibly scaling at maximum speed for much longer.” Full treatment: openai-alien-mind, jakub-pachocki.
• From the same source (@OpenAI, 2026-09-04): GPT-6 Astra available to all Pro, Enterprise, and Business Premium users in ChatGPT Work and Codex, and live in the API, with Plus/Business still pending; @sama then said Plus/Business were out. Updates the Sep 3 Trusted-Access / “coming days” filing. Do not re-litigate pricing or benchmarks. Full treatment: gpt-6-astra.

Research-acceleration / RSI metrics (issuer, September 6, 2026)

• From 2026-09-07-x-overnight-openai-rsi-research-acceleration-data-jensen-agi (issuer Research acceleration: The view inside OpenAI, dated September 6, 2026; clip curl 2026-09-07; live re-fetch this ingest 403 / JS wall): OpenAI-claimed mid-August snapshot — 3.1 agent-workdays per human workday in the research org; median researcher >$600/day inference; 90th percentile >$7,000/day tokens; “automated research intern” goal claimed met; March 2028 automated AI researcher target. Explicit: “We do not yet know how to safely get all the way to aligned, full RSI.” Staff X: @kliu128, @_Chris_Ong, @sama RT. Do not collapse into “RSI achieved.” 3.1× is runtime vs human labor, not output. Full treatment: openai-research-acceleration. Distinct from gpt-6-astra product access and from openai-frontier-rl-pause’s August 18 X thread (this post adds Jul 20 / Aug 7 dated grain).

Fluid-proofs credit dispute (contested, September 8, 2026)

• From 2026-09-08-x-morning-buckmaster-alpoge-ai-fluid-proofs-openai-credit (statement PDF + @SebastienBubeck): tristan-buckmaster alleges Sep 6 calls with sebastien-bubeck about a ~100-page forced Navier–Stokes blowup and authorship/announce proposals that would drop levent-alpoge; Bubeck called the allegations “false and inflammatory” and promised more “tomorrow.” Buckmaster’s own non-claims: has not seen OpenAI’s proof; is not accusing data misuse. Codex (esp. GPT-5.6 Sol) named as a tool on the Buckmaster–Alpöge program; Astra “only used for writeups and auditing.” Keep both sides. Do not flatten “OpenAI stole NS from Codex logs.” Full treatment: openai-fluid-proofs-credit-dispute, buckmaster-alpoge-forced-blowups. Not a rewrite of openai-research-acceleration / gpt-6-astra.

Forced Navier–Stokes C/D claim + Images 2.5 (afternoon, September 8, 2026)

• From 2026-09-08-x-afternoon-openai-navier-stokes-images-2-5 (On the Navier–Stokes Millennium Prize Problem + @OpenAI, ~1:20pm ET): issuer claim of an AI-produced forced NS solution framed as Clay/Fefferman C and D (smooth external force; energy finite; Lean). Internal model claimed ≫ gpt-6-astra; ~10,000 agents; ~88 hours; Lean ~17 hours via Astra. “We do not intend to claim the Millennium Prize.” Issuer claim / pending independent review. Not unforced A/B. Full treatment: openai-forced-navier-stokes-claim. Side-result unforced Euler: openai-unforced-euler.
• From the same source (issuer Concurrent work + @OpenAI + @SebastienBubeck + @sama): afternoon credit replies — congratulate forced-Euler priority; no specific user data; de-identified product-usage data not ruled out; Bubeck clarification vs morning PDF. Keep parallel accounts. Full treatment: openai-fluid-proofs-credit-dispute.
• From the same source (Introducing ChatGPT Images 2.5): chatgpt-images-2-5 — Sketch/templates/comment edits; API Flare / Sunburst. Dollar table not extracted. Feature X posts are video (flagged, not grain).

Overnight Astra Work/Codex rollout + GPT-TV; SMB plugins flagged (2026-09-08/09)

• From 2026-09-09-overnight-x-openai-astra-full-work-codex-rollout-aa-model (@OpenAI, 2026-09-08 21:06:10Z): “Astra is fully rolled out to Plus, Pro, Business, and Enterprise users in Codex and ChatGPT Work.” Points to openai.com/gpt-tv/. Full treatment: gpt-6-astra, gpt-tv. Do not invent GA. Chat-picker vs Work/Codex left open (secondary NotebookCheck says Plus may stay Work/Codex-only; overnight post does not mention the picker).
• From the same source (@OpenAI, 2026-09-09 03:12:23Z): “Meet the 16 plugins” for small business; link chatgpt.com/plugins?category=small-business. Video attached — flagged, not grain. Logged-out fetch was a thin “# Small Business” shell. Do not invent plugin names. @gdb quotes the same post — flag only. Wait for /transcribe-clipping before using as grain.

Paul Christiano joins Foundation Board + SSC (official, September 9, 2026 afternoon)

• From 2026-09-09-afternoon-x-anthropic-cyber-incident-alignment-assessment (@OpenAI, 2026-09-09 17:39:20Z + issuer page): paul-christiano (ARC founder; NIST CAISI Senior Tech Advisor) joins the OpenAI Foundation Board and its Safety and Security Committee, plus non-voting observer on the OpenAI Group PBC Board. Issuer footnote: will recuse from all OpenAI-related matters and all model evaluations. Dual-role optics left open. Full treatment: openai-christiano-foundation-board. Did not restate the May 2026 Foundation $130–260B figures. Distinct from overnight gpt-6-astra Work/Codex rollout.

Defense Factory playbook (official, September 9, 2026 overnight)

• From 2026-09-10-overnight-x-openai-defense-factory-uk-aisi-mythos-51-access (@OpenAI, 2026-09-09 20:37:58Z + Defense Factory): continuous agent-first find/validate/fix loop from an internal sprint (issuer: 250+ people, 100+ service areas, 53 urgent/high closed day one). Sprint metrics are workflow self-reports, not an independent audit. Merged ≠ deployed. Daybreak Blue/Red named as security models — not a rewrite of openai-astra-critical-cyber. Full treatment: openai-defense-factory. No attack procedures. Distinct from the afternoon openai-christiano-foundation-board appointment. Did not re-file Christiano / Astra Work-Codex permalinks.

ChatGPT Work Data agent (official, September 10, 2026 midday)

• From 2026-09-10-openai-data-agent-in-chatgpt-work-connects-warehouses-and-bi (Now everyone can put data to work, dated September 10, 2026 + @ChatGPT): Data plugin in ChatGPT Work — connect approved warehouses/files/BI, ask in natural language, investigate changes, share interactive dashboards without writing SQL. Queries “enforce the connected account’s existing permissions, including table, row, and column restrictions.” Issuer grain. VentureBeat (no published retrieval/accuracy benchmark; Shah “in their state” / no persistent customer context layer) is secondary. Announce video was transcribed (Whisper base) but output is unintelligible — not capability grain. Full treatment: chatgpt-work-data-agent. Not a rewrite of gpt-6-astra / overnight SMB 16 plugins / openai-defense-factory. Distinct from the 10am deepseek-v4-1-flash filing.

ChatGPT for Financial Services (official, September 10, 2026 afternoon)

• From 2026-09-10-openai-chatgpt-for-financial-services-pairs-astra (Introducing ChatGPT for Financial Services, dated September 10, 2026 + @OpenAI): tailored ChatGPT Work experience that “combines built-in financial data with GPT‑6 Astra’s reasoning” for research, financial models, and customized client materials. Design partners named: Morgan Stanley and Evercore (partner statements, not verified outcomes). Built-in / hosted (issuer): Daloopa, PitchBook, LSEG News, Crunchbase. Eligible financial institutions via contact/account team — not a consumer launch. Issuer grain. CNBC (Turley efficiency frame; unnamed customers; Claude for Financial Services as competition) and Reuters-via-WKZO (Quartr / connector mix) are secondary. Announce video was transcribed (Whisper base, 1m 21s) but output is . — not capability grain. Full treatment: chatgpt-for-financial-services. Not a rewrite of midday chatgpt-work-data-agent, gpt-6-astra (do not invent GA or re-litigate $10/$50), overnight openai-defense-factory, or 10am deepseek-v4-1-flash. Do not promote Sol as released.

Tensions / open questions

• Does OpenAI sustain consumer leadership while ceding the coding-agent layer? The Sax/Benioff/Turley framings collectively suggest Anthropic now leads coding-agent share. OpenAI has the consumer brand; Anthropic has the developer enterprise pull. Whether these are stable parallel franchises or one wins the other depends on whether super-assistant deployment generalizes off the consumer base into enterprise faster than Anthropic's enterprise-AI agent deployments generalize back to consumer.
• How does the pricing evolution land? Turley's "unlimited electricity plan" framing predicts metered pricing — but the timing and structure (per-token? per-hour-of-agent-time? per-agentic-task?) is unclear. The wiki should watch for SKU announcements.
• Apple resolution. If OpenAI loses Apple as a distribution partner (or sues, deteriorating further), what's the secondary distribution path? Reliance (India, via Gemini partnership) suggests OpenAI is already pursuing region-specific distribution outside the Apple flywheel.

Related

• anthropic — the competitor lab; competitive dynamics tracked symmetrically on both pages
• nick-turley — primary on-record OpenAI source for the page
• chatgpt-super-assistant-vision — Turley's product vision
• gpu-as-zero-sum-constraint — the operational constraint shaping product trade-offs
• ai-coding-productivity-paradox — Turley's "engineers don't open IDE" Mar 2026 corroboration sits here
• three-camps-of-ai — OpenAI is the canonical Camp 1 lab
• claude-opus-4-5, claude-code — the Anthropic side of the competition
• andrej-karpathy — OpenAI co-founder who left to do education work (Eureka Labs)
• ai-value-accrual-and-universal-basics — the OpenAI Foundation is the worked example
• paul-christiano — September 9, 2026 afternoon: ARC founder / NIST CAISI advisor appointed to Foundation Board + SSC
• openai-christiano-foundation-board — same appointment; issuer recusal footnote; dual-role question left open
• openai-defense-factory — September 9, 2026 overnight official: Defense Factory playbook; 250+ person sprint; issuer workflow metrics, not an audit
• chatgpt-work-data-agent — September 10, 2026 midday official: Data plugin in ChatGPT Work; issuer connectors/permissions/@Data; no published accuracy benchmark (VB secondary); video transcript not grain
• chatgpt-for-financial-services — September 10, 2026 afternoon official: tailored ChatGPT Work + hosted Daloopa/PitchBook/LSEG News/Crunchbase + gpt-6-astra; design partners Morgan Stanley and Evercore; eligible institutions only; video transcript not grain
• agi-definitions-and-benchmark-saturation — the GPT-5.5/Opus-4.8 convergence backdrop
• microsoft — the former partner now building competing first-party models
• ai-robotics-embodiment-wave — OpenAI's in-house robotics pivot
• openai-frontier-rl-pause — August 2026 official X: two-week RL pause; largest frontier RL run on hold
• openai-private-safety-processing — August 2026 official X: ZDR stays; Private Safety Processing preview (no personnel access to content)
• gpt-5-6-sol-pricing — August 2026 official X: GPT-5.6 Sol API/credit price drop of over 20% for 3 months (no dollar rate)
• openai-jalapeno — August 2026 official X: first custom inference chip; qualitative only; year-end deploy; no vs-Nvidia multiples
• chatgpt-business-premium-seats — August 2026 official X: $100 Business Premium seats; @gdb: no 5-hour limit
• openai-hugging-face-incident — August 2026 official X: HF investigation + technical report (agents' activity, failed safeguards, recurrence); METR/Redwood third-party assessment (report not fetched); HN headlines stay contested
• openai-astra-critical-cyber — September 1, 2026 official: first Preparedness Critical cyber designation; ExploitBench known-vuln 100% vs separate Internal Port zero-days; not GA
• gpt-6-astra — September 3, 2026 official: computer-use flagship SKU; Trusted Access today; $10/$50; not GA; marketing page not retrieved. Updated 2026-09-04: banked ChatGPT resets + messy-rollout apology + Azure Foundry Limited Access; still not GA; consumer vs Foundry channels kept split. Updated 2026-09-07: Sep 4 issuer text — Pro / Enterprise / Business Premium + API live; Plus/Business then out per @sama. Do not re-litigate $10/$50. Updated 2026-09-09: issuer “fully rolled out” in Codex + ChatGPT Work; chat-picker left open.
• gpt-tv — September 8/9 overnight: Astra-powered livestream/demo surface at openai.com/gpt-tv/; not a model card
• artificial-analysis-model-release-pages — September 9 AA product/UI (not an OpenAI card)
• openai-wiki-incident — September 5, 2026 official X: wiki incident + misalignment-incident disclosure framework (upcoming weeks). Reuters-attributed counts not asserted.
• openai-alien-mind — September 6, 2026 Pachocki essay (Altman-endorsed). RSI expectation; withhold-scaling language; Astra-vs-Sol alignment is issuer claim.
• jakub-pachocki — OpenAI Chief Scientist; essay author
• openai-research-acceleration — September 6, 2026 official: internal RSI / agent-workday / intern-goal metrics; full aligned RSI unsolved; Jul 20 / Aug 7 pause grain
• openai-fluid-proofs-credit-dispute — September 8, 2026: Buckmaster allegations vs Bubeck denial vs afternoon OpenAI/Bubeck/Altman replies; keep both sides
• openai-forced-navier-stokes-claim — September 8 afternoon: issuer forced NS C/D (smooth force); not claiming prize; pending independent review
• openai-unforced-euler — same clip: issuer unforced Euler side-result (~100 agents / ~50h)
• chatgpt-images-2-5 — September 8 afternoon: Images 2.5 + Flare/Sunburst; no invented prices
• sebastien-bubeck — named in that dispute
• buckmaster-alpoge-forced-blowups — public Lean-checked forced Euler/Boussinesq/IPM (not Clay NS)
• kiro — August 2026 @OpenAIDevs: GPT-5.6 available in Kiro (not @OpenAI)
• agent-harness
• agents-md-vs-claude-md
• terminal-bench-harness-swing
• codex-cli