Codex opens up to Kimi K3 and GLM-5.3 Flash via Baseten

Baseten has joined OpenAI’s B2B Marketplace as an inference provider for open models, and OpenAI enterprise customers can run GLM-5.3 Flash and Kimi K3 natively in Codex, the company said on September 29 in a post on X. If your company already has a spending commitment with OpenAI, its spend on those models through Baseten counts toward that commitment. Native access in Codex is waitlisted.
OpenAI’s marketplace page says the program lets customers “use part of your existing OpenAI commitment toward eligible partner products.” It lists Baseten as “the platform for production inference, execution, and post-training.” The commitment money covers two routes. One is Codex, OpenAI’s coding tool, where you keep the Codex interface and Baseten serves the open model behind it. The other is the Responses API, which developers call from their own applications. Neither company has published per-model prices.
In its announcement, Baseten pitches the deal at teams doing agentic coding. It says these teams send tasks “through static or adaptive routers, to the model that delivers the best mix of cost, quality, and performance for the job,” and that the gap between new open and closed model releases is “down to weeks.” Baseten says it launches major open models on its Model APIs on the day they ship, and that trying a new one is “as simple as updating a single line of code.” It also calls code generation a hard load to serve: “Multiple turns, long prompts over large repos, and huge context windows tax every part of the stack.”
The infrastructure figures are Baseten’s description of its own platform. It says it runs on more than 90 clusters across 20+ clouds and works with 200+ compute providers. It also runs active-active deployments, with live copies in several places at once, so losing a provider or a region reroutes traffic instead of stopping work. Inference runs on US-based infrastructure with zero data retention for all prompts, which means prompts are not stored. Deployments can be pinned to specific regions, access controls are fine-grained, and usage can be viewed by model, user or key. Blaxel, now part of Baseten, gives each agent its own sandbox. So far, the only concrete change for buyers is the billing. The rest has not been tested outside Baseten.
To get the open models in Codex under an OpenAI commitment, an enterprise joins Baseten’s waitlist. For other use cases, Baseten offers direct consultations with its engineers.
Up next
OpenAI hit with anti-hacking suit over Hugging Face breach
The nonprofit suing OpenAI over the Hugging Face breach says that, to its knowledge, it was never hit by an autonomous hack. It still wants a court order barring OpenAI’s tech from other companies’ systems.
Related stories
- Codex keeps coding in the cloud after your laptop closes
- Google Cloud gets OpenAI’s GPT-6 models in private preview
- Codex helped fix a 2007 Radeon video bug for Linux 7.5
- Prompt injection reached the official Codex npm pipeline
- OpenAI to buy Pierre, whose diff tools already run in Codex
- OpenAI’s Pando cut monorepo push latency nearly 5x
Comments
No comments yet. Be the first.
Join the conversation
Sign in to leave a comment. It appears after moderation.

