An anonymous X account posted on Saturday that OpenAI may have had “the worst week in the company’s history”, citing friends at the company. We checked the post’s eight factual claims against what has been reported.
Two hold up, three hold up in part and three could not be verified, by our count. Model scores for every lab are on our AI model benchmarks page.
The mood inside OpenAI, which the post says it heard from friends there, is not something we can check, and we have not tried to.
Business Pill: The Cost of Checking
An evergreen Business Engineer explainer on the cost of checking work. It teaches the general idea only and does not discuss OpenAI or this post.
The key insight: Of eight factual claims in a viral post about OpenAI’s week, two match the record, three match it in part and three could not be verified, by our count; its $1.5 trillion target and four-month runway are not on record.
Claim 1: Early Testers Panned GPT-6.2. Supported
OpenAI released GPT-6.2 Astra and Sol on Tuesday, 17 November. Of five named testers whose posts we read, four placed Astra below Anthropic’s models and the fifth did not recommend it.
Theo, who posts as Theo – t3.gg, wrote: “I think this is the worst model OpenAI has ever released”. He also called it a “great reviewer” and said “Computer use is nuts”.
The benchmarks are kinder. Epoch AI scored Astra at 174.17, level with Claude Fable 5.2 at 173.5. On the Artificial Analysis index Astra scored 62, behind Fable 5.2’s 65.

Claim 2: Theo’s Review Outdrew the Launch. Unverifiable
OpenAI’s launch post had 12,838,182 views when we read it on Saturday evening. We could not retrieve view counts for Theo’s posts, so we cannot compare them.
Claim 3: Google Overtook OpenAI in a Single Week. Partly
On Arena’s text leaderboard dated 20 November, Google holds the top three places, led by Gemini 4 Carbon at 1,527. Polymarket traders priced Google at 87% to own the best model at year-end, OpenAI at 2%.
But Gemini 4 Neon is third on that board, not first. Carbon has led it since 28 October, according to secondary reports we have not checked directly, which would be more than a single week.
Other scoreboards put OpenAI ahead of Google. Astra beats Carbon by 2 points on Artificial Analysis and 4.39 points on Epoch’s index, by our arithmetic. On Arena’s agent index, Anthropic’s Fable 5.2 is first and Neon second.
Claim 4: A New Nano Banana Model Is in the Arena. Unverifiable
The post says “Nano 🍌 Pro 2”, a banana emoji, is in the arena; Nano Banana is the name of Google’s image models. We found no report of a Nano Banana Pro 2 on Arena. TestingCatalog’s 19 November brief listed “Nano Banana 3 Pro, soon”.
Claim 5: Anthropic Solved the Riemann Hypothesis. Partly
The Lean FRO and the Mathlib Initiative said on Saturday: “We hereby certify that both the statement and the proof of the Riemann Hypothesis are correct.” Anthropic mathematician Levent Alpöge says he found the proof working with Claude.
That is a computer check, not yet acceptance. There is no human-readable paper yet, and the Clay prize rules require publication, two years and general acceptance.
On Navier-Stokes, OpenAI said in September that an internal model proved blow-up with an outside force, which many mathematicians viewed as a loophole, per OfficeChai. Claims of “bullying mathematicians” and how much compute OpenAI spent are not on record.
Claim 6: Wanted $30B at $1.5T, Got $15B at $1.25T. Partly
Reuters reported on 19 October that OpenAI was seeking about $30 billion at a valuation that could top $1.4 trillion. CNBC reported on 12 November that it had raised $15 billion at $1.25 trillion.
So the $30 billion and the $15 billion match reports, and $15 billion is half the earlier target, by our arithmetic. We found no report of a $1.5 trillion target.
Calling the round a disaster is the post’s view. CFO Sarah Friar described it as unplanned and “opportunistic”, Semafor reported, and $1.25 trillion is 46.7% above March’s $852 billion, by our arithmetic.
Claim 7: OpenAI Lost $10.9B Last Quarter. Supported
The Information reported on 17 November that OpenAI “lost nearly $11 billion last quarter on $10 billion in revenue”. That is roughly $1.10 lost per $1 of revenue, by our arithmetic. We did not find the exact $10.9 billion figure.
Claim 8: The Money Lasts About Four Months. Unverifiable
No cash balance for OpenAI has been reported, and a net loss is not the same as cash spent. Without both numbers, months of runway cannot be worked out.
The post’s prediction that the valuation will drop further is a forecast, not a fact, so we have not scored it.
The Structural Read
As we read it, the post mixes reported facts with numbers nobody has reported, and the reported ones carry it.
As we read it, who leads depends on the scoreboard: Google tops Arena’s text board, OpenAI is ahead of Google on two benchmark indexes.
Lean FRO and Mathlib Initiative, 21 Nov 2026
“We hereby certify that both the statement and the proof of the Riemann Hypothesis are correct.”
Three Implications
FOR READERS two of the post’s eight claims are fully on record.
FOR OPENAI the round, the loss and the reviews are reported; the runway is not.
FOR THE SCOREBOARD the leader changes with the index you read.
What Is Not Established
We could not read the post’s account of OpenAI staff, Theo’s view counts, a $1.5 trillion target or OpenAI’s cash balance anywhere on record. OpenAI has not commented on the post in anything we have seen.
The Bottom Line
The post’s strongest points match the record: testers panned GPT-6.2, Google leads Arena’s text board, and OpenAI reportedly lost nearly $11 billion last quarter. Its sharpest numbers, a $1.5 trillion target and four months of cash, are not on record.
97,000+ executives read Business Engineer for the AI strategy frameworks cited by ChatGPT, Claude, and Perplexity.
A note on sourcing. The post we checked is from an anonymous X account; we quote its claims and do not rely on its account of OpenAI staff. Every verdict rests on reporting by named outlets, company posts and public leaderboards, linked through our earlier coverage. The halves, gaps and percentages are our arithmetic, and the tally of verdicts is our count.
Sources: Anonymous X post checked in this piece, 21 Nov 2026 17:20 UTC · OpenAI on X, ‘GPT-6.2 Astra and Sol are now available’, 17 Nov 2026 · FourWeekMBA, testers’ posts and Artificial Analysis index, 18 Nov 2026 · FourWeekMBA, Epoch Capabilities Index scores, 17 Nov 2026 · FourWeekMBA, Arena text leaderboard (20 Nov) and Polymarket odds, 21 Nov 2026 · FourWeekMBA, Arena Agent Arena Index and TestingCatalog brief, 19 Nov 2026 · FourWeekMBA, Lean FRO and Mathlib Initiative statement, 21 Nov 2026 · FourWeekMBA, Navier-Stokes claim and OpenAI’s September result per OfficeChai, 21 Nov 2026 · FourWeekMBA, funding round reporting (Reuters, CNBC, WSJ, Semafor), 17 Nov 2026 · FourWeekMBA, The Information on OpenAI’s quarterly loss, 17 Nov 2026









