On 1 November Dan McAteer posted that GPT-6.2 Astra and Fable 5.5 were “HIGHLY likely to be released next week”, and that OpenAI and Anthropic would “wait until after the U.S. mid-terms elections on Tuesday, Nov. 4th, to avoid social disruption.” Polling day was actually Tuesday 3 November. It came and went, and five days later neither model is out.

As a reason to hold a launch, the vote explains a couple of days at most. The theory caught on because it gave everyone a date, and the AI For Humans hosts were still saying this week that the new models were coming “maybe because the election has happened here in America.” If the labs were only staying clear of polling day, the Wednesday after was the obvious time to ship. Instead, on Wednesday The Information reported that OpenAI could release a new Astra, a new Sol and a cyber model as soon as Friday, 6 November. Friday came, and OpenAI announced BioBasicsBench, a 100-task wet-lab benchmark, and the models didn’t come. The election’s longer reach is another matter, and I’ll come back to it.

The models certainly look close. Decrypt counted the clues: a gpt-6.2-astra label taking 19.8% of one account’s Codex usage on 1 November, a “Mythos 5.5” label in the hidden code of Anthropic’s Mythos page (which still lists version 5.1) the same day, and a claude-fable-5-5 entry that turned up in LiteLLM’s model catalogue on 13 October before being pulled. On 5 November people found Astra and Sol on the quotas page of the Google Cloud console. One of the AI For Humans hosts said Astra had been “fully at capacity” for him all week, and that Anthropic’s current models had gone “big dumb”, his theory being that compute gets moved off the model you’re using to test the next one. I’ve no idea whether that’s real. It may just be what waiting feels like. Either way, a model that’s answering real Codex requests and has quotas set up with a cloud partner is plainly built.

Built isn’t the same as cleared, though, and OpenAI has already shown the difference this autumn. GPT-6.1 Astra was scrapped on 28 September after OpenAI decided it didn’t meet the company’s safety standards, and nobody outside could have told that from the plumbing. So when people say these models were surely ready by now, my answer is that the servers were. Sign-off is a separate gate, and it’s the one OpenAI has already shut once on an Astra.

The current explanation has the same shape. On Saturday AGTP passed on a claim that “serious alignment and safety issues were found at the last minute” and that “a few more days of polishing” should get the models out. It rests on one unconfirmed post, and AGTP says plainly that OpenAI hasn’t said anything. The other story on X is a game of chicken, with each lab waiting for the other to ship so its own flagship doesn’t get one day of headlines before the rival’s lands on top. I half believe that one, mostly because it would explain a wait that has outlasted the election without anyone needing to have found a problem.

Where the midterms could really matter is Anthropic’s IPO. The vote made AI an electoral issue: CNBC reported that 56% of voters called AI and data centres a crucial or major factor, and 65% of those backed a Democrat for Congress. I went through the market fallout in Ask the Bond Market. Then on 5 November Reuters reported that the AI backlash could threaten the timing of the IPO, whose roadshow had been expected to start in the week of 9 November. A company selling shares wants its best model in the news that week. A slipping roadshow cuts both ways, since Anthropic could ship Fable 5.5 early to steady nerves, but I’d bet on it holding the model for the new date. That would be a delay the election really did cause, routed through the share sale rather than polling day. I can’t show the two are linked, and Anthropic hasn’t said a word about Fable 5.5.

My own calls from Friday were GPT-6.2 Astra and Sol before 13 November and the next Fable model in the last week of November. The OpenAI call now needs a launch within five days. I can’t weigh one unconfirmed post about a safety problem either way, so I’d put it at even odds, which is a long way from where I started on Friday. I’m leaving the Fable call alone, because a late roadshow, if anything, nudges Anthropic towards it.

Sources: