A report published September 30, 2026, described an enterprise route for using Moonshot AI’s Kimi K3 through OpenAI Codex, with Baseten providing inference. Under the reported arrangement, eligible enterprise use could count against existing OpenAI spending commitments without a separate supplier contract.
The reported enterprise route uses Baseten for inference
In this arrangement, Codex is the developer interface and Baseten supplies the computing that runs Kimi K3. The reported billing path links eligible use to customers’ existing OpenAI spending commitments.
This is separate from Moonshot AI’s direct API setup. It is also distinct from the Codex and API changes OpenAI announced at DevDay, covered in our earlier report.
Moonshot AI’s direct setup uses a Kimi API key
Moonshot AI documents a direct way to configure kimi-k3 in both Codex CLI and Codex Desktop. Codex connects to Moonshot AI’s endpoint, https://api.moonshot.ai/v1, using its native Responses API support; the configuration identifies that protocol with wire_api = "responses". Authentication uses a Kimi Open Platform API key supplied through the KIMI_API_KEY environment variable.
The guide sets model_context_window = 1048576, a configuration value it says matches Kimi K3’s 1M-token context window. Codex CLI and Desktop share the same user-level configuration, and the guide says to restart Desktop after changing it.
For this direct route, users authenticate with a Kimi API key through Moonshot AI’s endpoint. That setup is distinct from the separately reported enterprise billing arrangement.
Codex CLI supports text and image input, but it has no native video-input channel.
