OpenAI's GPT-5.5 Doubles Token Pricing, Endorses Codex CLI as Subscription Path

OpenAI has released GPT-5.5, initially limited to its Codex agent and a rolling rollout to paid ChatGPT subscribers. When the API opens, it will land at $5 per million input tokens and $30 per million output tokens — exactly twice GPT-5.4's $2.50/$15 rate. A higher-tier GPT-5.5 Pro variant is priced at $30/$180. The 2x gap is deliberate product segmentation, not a rounding error.

OpenAI's release notes explain the delayed API access: "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." GPT-5.4 will remain available at its current price. The positioning mirrors the Claude model stack: GPT-5.4 playing the Sonnet role — capable and cost-efficient — while GPT-5.5 occupies the Opus tier for workloads that justify the premium.

Before the API goes live, there is an officially endorsed shortcut. OpenAI's developer-relations lead Romain Huet stated on March 30th: "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. That's why Codex CLI and Codex app server are open source too." Peter Steinberger — creator of the OpenClaw agent harness, now at OpenAI — followed up: "OpenAI sub is officially supported." The mechanism routes requests through the same /backend-api/codex/responses endpoint that the open-source Codex CLI uses.

Anthropic recently blocked OpenClaw from routing through Anthropic subscription accounts. OpenAI responded by explicitly welcoming the practice for its platform, turning a competitor's enforcement action into a recruiting and ecosystem play. Third-party tooling built against the Codex endpoint now has explicit cover from OpenAI leadership — a different posture than a silent API quirk.

An organization already paying for ChatGPT Pro or Team subscriptions can access GPT-5.5 via the Codex CLI endpoint today, bypassing per-token API billing until the formal API launches. Researcher Simon Willison built and published llm-openai-via-codex, a plugin for the LLM command-line tool that automates authentication by reading tokens stored by the Codex CLI. The install path is four commands: install Codex CLI, authenticate, install the LLM plugin, then prompt directly against gpt-5.5 or gpt-5.5 with elevated reasoning effort via the -o reasoning_effort xhigh flag.

Reasoning effort is not cosmetic. In Willison's SVG generation benchmark, default reasoning consumed 39 tokens while xhigh consumed 9,322 — a 239x difference — with visibly distinct output quality. Teams running agentic coding or complex multi-step reasoning pipelines on GPT-5.5 should budget for output token variance driven by reasoning depth, not just prompt length. For workloads that rely on extended thinking, that variability will matter more than the 2x base-price increase.

Researcher Ethan Mollick tested GPT-5.5 and GPT-5.5 Pro across challenging tasks and concluded the "jagged frontier continues to hold" — the model excels in certain domains while falling short in others in ways that remain hard to predict. That pattern is consistent with every prior frontier release and argues for empirical benchmarking on task-specific workloads before committing to a migration from GPT-5.4.

The Codex subscription path closes once the formal API launches "very soon," at which point per-token billing takes over and the 2x cost differential becomes real. Teams that want to run production evaluations, tune system prompts, or assess reasoning-effort scaling have a narrow, officially sanctioned window to do so at subscription cost.