GPT-5.5: OpenAI Enters a New Era with Codex and Its Challenges

⚡

Key Takeaways

1GPT-5.5 is now accessible via OpenAI Codex, with a gradual rollout for ChatGPT subscribers.

2OpenClaw, a controversial tool, has been blocked by Anthropic, but OpenAI welcomes it to integrate with its subscriptions.

3A new plugin, llm-openai-via-codex, allows leveraging Codex subscriptions to execute LLM queries.

💡Why it matters — The integration of GPT-5.5 into Codex and the openness to tools like OpenClaw illustrate OpenAI's strategy to democratize access to its models while managing security concerns.

GPT-5.5: A Major Advancement in the OpenAI Ecosystem

The GPT-5.5 model, recently launched by OpenAI, is now accessible through the Codex platform. This rollout is currently underway for paid users of ChatGPT. Having had the opportunity to test this model in advance, I was able to witness its remarkable speed and efficiency. It stands out for its ability to respond accurately to formulated queries, although describing its performance in words remains a challenge.

However, a notable absence persists: the API for GPT-5.5 is not yet available. OpenAI is actively working with its partners to ensure the necessary security for a large-scale deployment. The company promises that the API for GPT-5.5 and its Pro version will be accessible soon.

The Challenges of Integration with OpenClaw

In the current AI landscape, the interaction between agent environments like OpenClaw and the APIs of major providers is a topic of debate. OpenAI and Anthropic offer monthly subscriptions that allow access to their models at a lower cost compared to direct API access. OpenClaw has integrated into this system but has been blocked by Anthropic, leading to a series of reactions.

Seeing an opportunity, OpenAI welcomed OpenClaw to continue integrating into its subscriptions via the same mechanism as their open-source Codex CLI tool. OpenAI recently hired Peter Steinberger, the creator of OpenClaw, thereby strengthening their collaboration. This raises the question of whether any developer can create solutions that connect to OpenAI's Codex APIs to access these subscriptions.

The Codex Initiative and API Openness

Jeremy Howard recently questioned the community about OpenAI's official support for the endpoint /backend-api/codex/responses, used by tools like Pi and Opencode. On March 30, Romain Huet from OpenAI clarified the situation on Twitter, stating that the goal is to enable the use of Codex and ChatGPT subscriptions in various environments, including JetBrains, Xcode, OpenCode, Pi, and Claude Code.

Peter Steinberger confirmed that the OpenAI subscription is officially supported, paving the way for new integrations and uses of OpenAI models. Codex CLI and the Codex application server are also open source, facilitating access and innovation around these tools.

The llm-openai-via-codex Plugin: A New Era for LLMs

To explore these new possibilities, I asked Claude Code to reverse-engineer the openai/codex repository to understand the storage of authentication tokens. This led to the creation of llm-openai-via-codex, a plugin that allows the use of an existing Codex subscription to execute LLM queries.

To use this plugin, simply install Codex CLI, subscribe to an OpenAI plan, and then connect to Codex. Next, install LLM and the llm-openai-via-codex plugin to start making queries with GPT-5.5.

Creative Experiments with GPT-5.5

To test the creative capabilities of GPT-5.5, I attempted to generate an SVG of a pelican on a bicycle. Although the initial result was disappointing, adding the option -o reasoning_effort xhigh significantly improved the quality of the generation. This approach required more time and resources, but the final result was markedly superior.

Comparing the generated SVG codes, the xhigh approach used 9,322 reasoning tokens, compared to only 39 for the default version, thus demonstrating the significant impact of generation parameters on the quality of results.