GPT-5.5 Pricing: Full Breakdown of API, Codex, and ChatGPT Costs (April 2026)

OpenAI doubled the per-token price on the GPT-5 line with the April 23, 2026 release of GPT-5.5. Input goes from $2.50 to $5.00 per million tokens. Output goes from $15.00 to $30.00 per million. Pro pricing stays flat at $30 / $180. That is the headline; the details are where the actual cost lives.

This guide covers every pricing surface: API standard, Batch, Flex, and Priority; Pro pricing; Codex per-plan limits; and how to do the math on your real workload before you commit to a default model change.

For the model overview, see What is GPT-5.5. For the developer walkthrough, see How to use the GPT-5.5 API.

TL;DR

Net effect: GPT-5.5 is 2× GPT-5.4 at the token level, but OpenAI claims a ~20 % net Intelligence-Index increase once token efficiency is factored in.

The headline numbers

OpenAI published the pricing on the API pricing page the same day as the launch.

• GPT-5.5: $5.00 per million input tokens, $30.00 per million output tokens.
• GPT-5.5 Pro: $30.00 per million input tokens, $180.00 per million output tokens.
• Context window: 1 M tokens on both variants. Reasoning tokens count against the window and against output billing.

Batch, Flex, and Priority

OpenAI offers three alternative tiers that re-shape the standard price.

Batch API

Queue requests through the Batch endpoint and they run at 50 % of standard pricing. Turnaround is under 24 hours. Good for:

• Overnight evaluations on a full dataset.
• Backfills and historical re-processing.
• Any workflow where the latency budget is measured in hours, not seconds.

At Batch pricing, GPT-5.5 costs $2.50 / $15.00 per million tokens; identical to GPT-5.4 standard pricing. For offline workloads, the price doubling disappears.

Flex processing

Flex also gives you 50 % off standard rates, but the wait time is variable; anywhere from seconds to several minutes depending on load. Use Flex when you can tolerate unpredictable latency and want Batch-level pricing with synchronous-ish responses.

Priority processing

Priority costs 2.5× the standard rate ($12.50 / $75.00 per million tokens on GPT-5.5) and gives you faster-than-default throughput, higher rate-limit ceilings, and near-zero queue time. Reserve it for live user-facing experiences where tail latency shows up in retention metrics.

Thinking mode cost math

GPT-5.5 Thinking is the same model ID at a higher reasoning.effort. It does not change the per-token price; it changes how many tokens a single request uses. Expect the multiplier to fall in three bands.

A single xhigh call on a long prompt can easily run 20 K reasoning tokens; at $30 per million that is $0.60 for the reasoning alone, on top of whatever the final output tokens cost.

Budget per workload, not per request.

Codex pricing

Codex access is tied to the ChatGPT plan, not to per-token billing. On April 23, 2026 the structure looks like this.

For users who spend most of their time inside a terminal coding workflow, Plus or Pro is the cheapest way to run GPT-5.5; the flat monthly fee beats even Batch pricing once you cross a few hundred thousand tokens a day. The free path guide covers the no-cost entry point.

If these headline numbers push you toward the Anthropic side of the line, our companion guide on How to Use Claude Opus 4.7 for Free documents the no-cost paths worth comparing against GPT-5.5 list pricing.

Comparison: GPT-5.5 vs the rest of the line

When do you pay for GPT-5.5 and when do you keep GPT-5.4 or GPT-5.4-mini? The cost math depends on how output-heavy your workload is.

A rough decision flow:

• High-volume, low-risk output (classification, summarization, simple chat): GPT-5.4-mini.
• General production traffic where 5.4 already passes quality bars: GPT-5.4.
• Hard coding, multi-step agentic work, research chains: GPT-5.5.
• Correctness-critical output where a wrong answer is expensive to catch downstream: GPT-5.5 Pro.

A worked example: coding agent cost per task

A typical agentic coding session through the API on GPT-5.5 with reasoning.effort: "medium" looks like this.

• Input tokens per task (repo context + user prompt): ~15,000
• Output tokens per task (code + explanations): ~3,000
• Reasoning tokens per task (medium effort): ~6,000

Cost per task on standard pricing:

• Input: 15 K × $5.00 / M = $0.075
• Output: (3 K + 6 K) × $30.00 / M = $0.27
• Total: $0.345 per completed coding task.

Running the same workload on GPT-5.4:

• Input: 15 K × $2.50 / M = $0.0375
• Output: 9 K × $15.00 / M = $0.135
• Total: $0.1725 per completed coding task.

GPT-5.5 is exactly 2× the cost per task at the same reasoning effort. The upgrade pays for itself when the SWE-bench gap (88.7 % vs roughly 74 %) closes enough tasks successfully to eliminate a second round-trip. If one-in-eight tasks on GPT-5.4 needs a manual rewrite, GPT-5.5 is cheaper in aggregate.

Cost controls worth building in Day 1

Five levers that keep GPT-5.5 bills boring.

• Hard max_output_tokens caps. Every call, every time. Default to 2,000 unless you explicitly need long output.
• Strict JSON schemas. Malformed output means retries; retries are a full priced call.
• Route by difficulty. Triage easy requests to GPT-5.4-mini; escalate hard ones to GPT-5.5. A 10-line router saves more than any prompt-level optimization.
• Use Batch for anything offline. Evaluations, backfills, nightly report generation; all 50 % off.
• Track usage.reasoning_tokens. The billing surprise on GPT-5.5 is almost always reasoning-token spend at high effort. Alert on it.

Per-plan monthly cost estimate

If you are picking a ChatGPT tier for GPT-5.5 access, here is what each plan runs.

For anyone running more than ~4 M output tokens per month on the API, Pro in ChatGPT plus the Codex CLI ends up cheaper than pay-as-you-go API billing, as long as the workload fits inside the CLI’s 400 K context window.

Price change signals to watch

Two signals worth tracking if you are making a long-horizon budget.

• The GA of the GPT-5.5 API. Pricing could shift down as OpenAI responds to competitive pressure from Anthropic Claude Mythos, Gemini 3.5, and open-weight models benchmarked in the Vellum leaderboard.
• Pro model democratization. OpenAI has historically dropped Pro-tier prices within 3–6 months of each release. Do not assume today’s $30 / $180 is permanent; do not assume it will drop either.

FAQ

Does caching reduce input cost?Yes. Cached input tokens on GPT-5.5 are billed at a fraction of the standard rate; the OpenAI pricing page has the exact multiplier. Cache anything you re-use in more than one request (system prompts, tool schemas, repo context).

Is there a volume discount?Not a published one. Enterprise contracts often include custom rates, and OpenAI adjusts pricing in response to sustained usage. Talk to sales past seven figures of annual spend.

Does thinking mode cost extra on top of per-token pricing?No. It costs more because it uses more tokens, not because the per-token rate changes.

Is Codex CLI usage billed separately from API usage?Only if you sign in with an API key. ChatGPT sign-ins bill the plan fee; API-key sign-ins bill the usage-based account.

What is the cheapest way to try GPT-5.5?Free or Go plan plus Codex CLI. See our free path guide for every zero-cost option.