OpenAI will reportedly ship at least one model this year but its flagship still needs alignment work

OpenAI will likely release one or two new models before the end of the year, according to employees cited by The Information, though it is unclear whether the major upgraded flagship the company is working on will be among them. The flagship still needs alignment work so its goals match what OpenAI's researchers and customers expect, and the company must also shore up the testing infrastructure around the model before it can move forward.
The details come from reporting by The Information, which cites people familiar with the projects. The article was republished by The Decoder.
After leadership is satisfied with safety evaluations, the flagship faces another hurdle: under an executive order from President Trump governing evaluations of advanced AI systems, OpenAI agreed to give the US government access to new models up to 30 days before they go to outside partners, and longer still before a public launch.
The next flagship uses the same looped transformer technique as Astra, OpenAI's current top model, according to a person familiar with the situation. OpenAI already runs this technique in production under the name Recurrent Depth. Rather than passing a token through the transformer layers once, the model loops it through the same layers multiple times before outputting the next word. That approach boosts performance on math and coding and lowers costs, because a smaller model can punch above its weight.
The tradeoff is that looping moves some of the model's thinking out of readable text and into internal numerical representations that human reviewers cannot inspect. That undermines chain-of-thought monitoring, a safety practice in which researchers trace a model's reasoning to spot bad behavior. OpenAI deliberately capped how many loops Astra runs to keep its chain of thought readable, according to The Information.
After an incident involving Hugging Face, OpenAI moved roughly ten percent of its total compute toward alignment and safety, according to a person familiar with the figures. The Information reports that amounts to hundreds of megawatts worth of data center capacity, though it is unclear how much was allocated to safety before or what the current split looks like. At market rates, the compute OpenAI shifted would cost billions of dollars a year to rent. AI researcher Noam Brown said in a mid-September podcast that about ten percent of his team also works on security and safety.
How OpenAI actually uses the technique became public through a slip-up at Microsoft. The major investor, which has access to OpenAI's intellectual property, published a technical write-up on GPT-6.1 Sol, a cheaper Astra alternative that OpenAI launched earlier this month. In a now-deleted section, Microsoft described GPT-6.1 Sol as using looping inference, where the model works through the same problem in multiple passes before producing an answer. GPT-6.1 Sol uses the same base model weights as GPT-6 Sol with two inference passes instead of three, according to a cached copy of the page quoted by The Information. A person with direct knowledge confirmed the write-up matched what OpenAI had told Microsoft in private. Microsoft took the section down after realizing it was not meant to be public.
OpenAI chief scientist Jakub Pachocki has acknowledged that chain-of-thought monitoring is growing fragile and, in his words, unfortunately trending in a negative direction. He has also warned against an industry-wide race toward models without readable reasoning.
The picture that emerges is of a company juggling capability gains against oversight. Looped or recurrent-depth transformers let a model spend more internal computation on a problem without adding parameters, which is attractive for math and coding workloads where accuracy matters and inference cost is a constraint. But if that extra computation happens in a form reviewers cannot read, the standard safety technique of inspecting a model's reasoning becomes less effective. This suggests OpenAI's decision to cap loops on Astra is a deliberate compromise, trading some performance for a chain of thought that remains legible. Whether the next flagship can make the same compromise without losing the gains that motivate the technique is the open engineering question.
Why it matters: For developers and product teams planning around OpenAI's roadmap, the near-term signal is that one or two models are likely to arrive this year, but the flagship may not be one of them. The reported 30-day government pre-access requirement adds a scheduling constraint that could delay partner and public availability even after internal evaluations pass. Teams that depend on chain-of-thought transparency for their own evaluations should note the tension between looped architectures and readable reasoning, since the same tradeoff may surface in the models they adopt next. This is inference from the reported facts, not a stated OpenAI plan.
Based on reporting from the original publisher. Visit the source for full context and later updates.
Publisher excerpt
OpenAI is racing to finish its next flagship model, which uses the same looped transformer technique as Astra to boost math and coding performance. The tradeoff is that looping makes safety monitoring harder. Microsoft accidentally revealed details about how the technique works on a public page. Whether or not the flagship is ready in time, employees say one or two new models will likely ship by year's end. The article OpenAI will reportedly ship at least one model this year but its flagship still needs alignment w