OpenAI Python SDK v3.18.0 Adds Identifiers for 'GPT-6 Sol' and 'GPT-6 Luna'

OpenAI shipped version 3.18.0 of its official Python SDK on September 22, adding code references to two previously unannounced model identifiers: "GPT-6 Sol" and "GPT-6 Luna."

The change is documented in a single line of the release notes: "api: add GPT-6 Sol and Luna model identifiers" (commit 455ce1b). No accompanying blog post, model card, pricing sheet, or benchmark data was published alongside the SDK update.

What we know

The openai-python repository is OpenAI's official client library, used by developers to call the company's API from Python applications. SDK releases routinely add support for new endpoints, parameters, and model strings ahead of or alongside public announcements. Adding an identifier to the SDK typically means the corresponding API endpoint exists or is being prepared server-side, but it does not confirm a public launch, availability, or final naming.

No context window size, pricing per token, parameter count, or benchmark scores have been disclosed for either "Sol" or "Luna." OpenAI has not issued any statement confirming these as production models, and the names could represent internal codenames, preview builds, or entries later renamed before public release.

What this means

SDK commits that add model identifiers are one of the more reliable early signals in the AI industry — companies generally don't wire up client libraries for names that will never ship. The appearance of two identifiers under a "GPT-6" naming scheme suggests OpenAI is actively preparing next-generation models beyond the GPT-5 line, with "Sol" and "Luna" likely denoting distinct variants (potentially differentiated by capability tier, speed, or reasoning mode, following the pattern of prior GPT-5 mini/nano splits).

However, developers and reporters should treat this strictly as a changelog artifact, not a launch. There is no confirmation of release timing, capabilities, or pricing. Attempting to call these identifiers via the API today will likely fail unless OpenAI has already enabled limited access. Until OpenAI publishes an official announcement or model card, all details beyond the existence of these two strings in the SDK remain speculative. Expect a formal reveal, if one comes, to include the standard disclosures: context window, input/output pricing per million tokens, benchmark results, and a training data cutoff date.