Two unannounced "rc-ada" models appear in OpenAI's API and are quickly removed; possibly voice models, but views on their use differ (leak)

Two names in the API list, soon removed

Two unannounced names, "rc-ada-or-vm" and "rc-ada-sh-vm," appeared in the model list of OpenAI's developer API on Nov. 3 (early Nov. 4 Japan time) and soon disappeared. @gptbotio, an X account that automatically detects and posts changes to the API list, reported the appearance of "rc-ada-or-vm" at 3:41 a.m. Nov. 4 Japan time. After disappearing from the list, the models cannot be called, and there is no documentation or pricing for them. OpenAI has made no announcement.

South Korean AI news site promppy also reported on Nov. 4 that the two names had briefly appeared in a model tracking tool and then disappeared. As far as ORIENS confirmed on Nov. 4, the name rc-ada does not appear on OpenAI's official API pricing page.

Views differ on the meaning of "vm"

Guesses about the meaning of the names differ. promppy reported an industry view that "rc" stands for "release candidate," a software version just before release, and "vm" marks a version that runs computer operations in an isolated environment (a virtual machine). No explanation has been found of what "or" and "sh" refer to.

On X, there is also speculation that "vm" stands for voice model. Both views are guesses based on the names, and OpenAI has not disclosed their use.

promppy's basis is an AI-generated summary of posts on the U.S. forum Reddit, and the site notes that it may be inaccurate. The article writes "rc-ada-vm" in its headline but lists two names in the body, and parts of its account are inconsistent. The article placed the names alongside talk of "GPT-6.2 Astra," rumored to be the next top model, and cited voices expecting a release around Thursday (Nov. 5), but gave no evidence linking it to rc-ada.

"ada" is also a name OpenAI has used before. The smallest of the older GPT-3 base models was "ada," and OpenAI ended its availability in the API on Jan. 4, 2024.

If voice models, current pricing is the benchmark

If rc-ada turns out to be voice models, the comparison would be with the current real-time voice models. According to OpenAI's pricing page, "gpt-realtime-2.1," released in July, costs $32 per million tokens for audio input and $64 for output. The smaller "gpt-realtime-2.1-mini" costs $10 for input and $20 for output. "gpt-live-1," specialized for conversations in which it speaks while listening, charges $0.05 per minute based on conversation time.

OpenAI is replacing its voice models. On July 20, it notified developers that it would remove older audio, real-time and transcription models from the API on Jan. 20, 2027, naming gpt-realtime-2.1 and others as successors. On Oct. 1, it announced that three models including the speech synthesis model "tts-1" would end on Jan. 6, 2027, and on Aug. 26 that the transcription model "whisper-1" and others would end on Feb. 26, 2027. Companies using these models must move to successors by the deadlines.

In ChatGPT, "GPT-Live-1.1" for conversation and the smaller "GPT-Live-1.1 Quip" were released to all users on Oct. 15. According to CellCog, OpenAI has said 1.1 will be available in the API "soon," and the API still uses "GPT-Live-1."

Unannounced names keep appearing before announcements

Since October, names of unannounced OpenAI models have repeatedly been found before official announcements. On Oct. 18, three "aurora" models, believed to be image generation models, appeared in the list and soon disappeared, and were later reported to be under testing on an evaluation site. On Oct. 31, "gpt-6-astra-law," aimed at legal work, appeared in the list and was added to the pricing page the same day. On Nov. 1, "gpt-6.2-astra" appeared in the logs of some users of the development assistance service "Codex." In each case, OpenAI had made no prior announcement.