OpenAI Launches GPT-5.5 Without API Access at Double GPT-5.4 Pricing

GPT-5.5 launched April 23 inside OpenAI Codex and is rolling out to paid ChatGPT subscribers, but API access is withheld indefinitely. OpenAI cites scale-safety requirements and has priced the eventual API at $5 per million input tokens and $30 per million output tokens — double GPT-5.4's rates.

In today's release, OpenAI said: "API deployments require different safeguards and we are working closely with partners and customers on the safety and security requirements for serving it at scale. We'll bring GPT-5.5 and GPT-5.5 Pro to the API very soon." The "very soon" framing carries no concrete date, leaving enterprise teams that rely on programmatic access without a timeline.

GPT-5.4 costs $2.50 per million input tokens and $15 per million output tokens. GPT-5.5 doubles both. The Pro variant goes further — $30 per million input tokens and $180 per million output tokens — positioning it as the Opus to GPT-5.4's Sonnet in OpenAI's emerging model-tier logic. GPT-5.4 remains available at its current rates.

With the API unavailable, a semi-official alternative is already active. OpenAI's /backend-api/codex/responses endpoint — the same one used by the open-source Codex CLI — has been publicly endorsed for third-party integrations by OpenAI developer relations head Romain Huet. On March 30, Huet wrote: "We want people to be able to use Codex, and their ChatGPT subscription, wherever they like! That means in the app, in the terminal, but also in JetBrains, Xcode, OpenCode, Pi, and now Claude Code." OpenClaw creator Peter Steinberger, now an OpenAI employee, confirmed the position: "OpenAI sub is officially supported."

Any developer with an active ChatGPT or Codex subscription can route prompts to GPT-5.5 today. Tooling author Simon Willison demonstrated this within hours of launch, publishing llm-openai-via-codex — a plugin for his open-source LLM framework that authenticates against the Codex CLI token store and proxies requests to the endpoint. The plugin exposes the full LLM feature set: image attachments, multi-turn chat, tool calls, and access to GPT-5.5's configurable reasoning modes.

Anthropic blocked OpenClaw — an agent harness that integrated with Anthropic's subscription tier — earlier this year. OpenAI's public endorsement of the Codex endpoint as an approved integration path is a direct contrast. By naming Claude Code as a blessed Codex integrator, OpenAI sharpened the irony: Anthropic's own coding tool can legally ride OpenAI's subscription rails. OpenAI subsequently hired Steinberger, the developer Anthropic had antagonized.

The critical question for enterprise architects is whether the Codex endpoint offers production-grade reliability. It does not. OpenAI has published no SLAs, rate limits, or versioning commitments for the backend endpoint — it is open-source infrastructure that OpenAI has declined to block, not a supported product. Teams evaluating GPT-5.5 for production workloads should treat the Codex path as sandbox-grade access until the formal API ships.

Reasoning costs scale sharply at higher effort settings. In Willison's testing, the xhigh reasoning effort level consumed 9,322 reasoning tokens on a single SVG generation task, versus 39 tokens at the default setting — a 239x difference. At $30 per million output tokens, sustained xhigh-reasoning workloads will register quickly on enterprise spend dashboards.

OpenAI's move recasts how AI labs handle the subscription-versus-API divide. The edge went not to the lab with the better access policy, but to the one that shipped the open-source client first.