Codex is now configured to route requests through a pluggable model access layer that can direct traffic to OpenAI models, local models, or third-party services via a base URL, wire API, environment key, and model mappings.

OpenAI Codex adds support for open-source models through the pluggable access layer, without changing the core GPT model itself.

A new OSS mode enables Codex to connect to local open-source model services like Ollama and LM Studio, allowing offline or local inference.

Hybrid routing approaches translate between APIs—such as bridging OpenAI’s Responses API with Chat Completions—to make OpenAI-compatible solutions like DeepSeek work with Codex.

Community response is mixed: users welcome openness and the cost/privacy benefits, but integration can be hampered by protocol differences between the Responses API and open-source interfaces.

A common use case combines GPT-based planning with offline or private open-source executors to cut costs and keep data local for developers who favor offline workflows.

Practical constraints remain: not all open-source models are instantly compatible, and there is no official endorsement of community routing tools yet.

These hybrid routing solutions can enable a broader ecosystem by letting Codex interoperate with non-native models through translation layers.

OpenAI’s move reconfigures the competitive landscape from model supremacy toward platform-level control and ecosystem entry points, potentially shaping future AI development dynamics.