OpenAI could release new AI models as early as Friday, November 6, 2026, including updated Astra and Sol models, according to a report by The Information. The possible launch comes as OpenAI has revealed an unreleased internal model that significantly outperforms GPT-6 Astra in engineering tests.
On November 3, OpenAI published benchmark results showing that its unreleased internal model achieved an average score of 87.5% on BenchCAD 2.0, compared with 59.4% for GPT-6 Astra at its maximum reasoning setting. BenchCAD 2.0 measures AI performance on engineering tasks involving 3D models, assemblies, and circuit boards.
An X user, @deredleritt3r, shared a more detailed comparison of the models’ performance and estimated costs across different reasoning settings on November 4, 2026. According to the post, even the internal model’s lowest estimated setting scores approximately 76%, significantly above Astra’s maximum score of 59%.
The X user provided the following estimated costs and scores per task:
Astra (low): $0.44 per task; score = 14%
Astra (medium): $2.73 per task; score = 31%
Astra (high): $5.11 per task; score = 41%
Astra (xhigh): $6.67 per task; score = 45%
Astra (max): $11.05 per task; score = 59%
Internal model (low?): ~$11 per task; score = ~76%
Internal model (medium, estimated): ~$23 per task; score = ~83%
Internal model (high, estimated): ~$34 per task; score = ~86%
Internal model (xhigh, estimated): ~$37 per task; score = ~88%
The X user said the internal model’s costs were extrapolated from OpenAI’s benchmark graph, while the reasoning-setting labels were estimates. These figures represent estimated costs of completing benchmark tasks, not confirmed API pricing. These figures also do not establish that the model outperforms Astra or Sol on every type of task.
The same X post claims OpenAI is preparing another Sol model that will benefit from the training methods used for its more capable internal model. According to the claim, the upcoming Sol model will not simply be a distilled version of the internal system. Instead, it will reportedly undergo similar training using games and intensive simulated environments designed to improve reasoning and problem-solving. The post suggests these methods could deliver substantial improvements over GPT-6.1 Sol.
According to The Information’s November 4 report, OpenAI could introduce new Astra, Sol, and cybersecurity-focused AI models as early as November 6. The X post goes further, claiming that the powerful internal model featured in OpenAI’s engineering research will also become available on Friday. However, The Information has not confirmed that the benchmarked internal model is among the upcoming releases. OpenAI has also not announced the new models’ official names, pricing, or availability. For now, the Friday launch remains unconfirmed.
