Editorial illustration for OpenAI reportedly to ship at least one model; flagship needs alignment
OpenAI's Flagship Model Delayed; New Models Coming Soon
OpenAI reportedly to ship at least one model; flagship needs alignment
OpenAI's next flagship model isn't ready, and the company's own staff aren't sure when it will be. The Information reports, citing people familiar with the projects, that OpenAI expects to ship "one or two new models" before the end of the year. Whether the big upgrade currently in the works makes that cutoff is still an open question.
The holdup isn't raw capability. It's alignment, the process of making sure a model's outputs actually match what OpenAI's researchers and customers want from it, plus the testing infrastructure needed to check that work before launch. There's also a regulatory clock running: under an executive order from President Trump covering evaluations of advanced AI systems, OpenAI has agreed to give the US government a look at new models up to 30 days before outside partners get access, and even longer before anything goes public.
Part of what's slowing things down is architectural. The new flagship reportedly uses the same looped transformer technique, internally called "Recurrent Depth," that already runs in production inside OpenAI's current top model, Astra.
OpenAI is racing to finish its next flagship model, which uses the same looped transformer technique as Astra. The approach boosts math and coding performance but makes safety monitoring harder. Employees say one or two new models will likely ship by year's end.
Why this matters
For anyone building on top of OpenAI's stack, the looped transformer tradeoff is worth watching closely. Faster math and coding gains are useful, but harder safety monitoring isn't a footnote, it's the thing standing between "shipped by year's end" and "shipped, period." If employees are already flagging that alignment work and testing infrastructure aren't done, that's a signal the timeline could slip again, something this company has a track record of. The executive-order hurdle mentioned but not detailed adds another layer of uncertainty that developers planning around a new flagship should factor in now, not after an announcement.
Founders building products on the assumption of near-term frontier access should have a fallback plan. Researchers should pay attention to how OpenAI talks about monitoring once the model ships, because a technique that trades interpretability for performance is exactly the kind of tradeoff that shows up later as an incident report. One or two models by December is a plan, not a guarantee.
Related Reading
Browse all OpenAI news →
Open Source D&B rebuilds 642 million‑business database after AI agents hit limits
Open Source Box CEO Aaron Levie Says AI Will Drive "Biggest Cybersecurity Upgrade
Open Source Study: 6.7% of Deepfake Requests on Hugging Face Targeted Children
Related Topic OpenAI researcher quits, citing distrust over ad‑driven engagement metrics
LLMs & Generative AI
Related Topic OpenAI launches GPT-Image 1.5 with precise editing for enterprise visuals
LLMs & Generative AI
From Archives Liquid AI releases LFM2.5-24B-A2B, a 2B-active MoE for on-device agents
From Archives Researchers Warn AI Guardrails Can Be 'Completely Removed
From Archives OpenAI, Google AI models easily tricked into making nudes, researchers say
Research & Benchmarks
From Archives OpenAI employee allegedly used chatbot to plant recording device at office
AI Tools & Apps
Common Questions Answered
Why isn't OpenAI's next flagship model ready for release despite having the capability?
The holdup is alignment, which is the process of ensuring a model's outputs match what OpenAI's researchers and customers want from it. Raw capability isn't the issue; rather, OpenAI needs to complete safety monitoring and alignment work before shipping the model.
What is the looped transformer technique used in OpenAI's flagship model, and what are its tradeoffs?
The looped transformer technique is the same approach used in Astra and boosts math and coding performance significantly. However, this approach makes safety monitoring harder, which is why alignment work has become the primary bottleneck for the model's release.
How many new models does OpenAI expect to ship before the end of the year?
According to The Information, OpenAI expects to ship one or two new models before year-end. However, whether the flagship model currently in development makes this cutoff remains an open question due to ongoing alignment challenges.
What risk does the looped transformer tradeoff pose for OpenAI's timeline?
The harder safety monitoring required by the looped transformer approach stands between shipping by year-end and shipping at all. If employees are already flagging that alignment work and testing infrastructure aren't complete, this could signal another timeline slip, particularly given OpenAI's track record of delays.
Further Reading
- OpenAI scrambles to fix safety issues on next flagship model - AI Weekly
- Pacing model development in an era of cyber-critical capabilities - OpenAI
- Safety and alignment in an era of long-horizon models - OpenAI
- Safety overview: GPT-6 Astra - OpenAI
- GPT-6 Astra: A new generation of intelligence - OpenAI
