Model comparison
GPT-5.5 vs GPT-5.4 on the API: price, caching, retirement
GPT-5.5 costs exactly twice GPT-5.4 on every rate, and both are leaving Codex sign-in. What the 2x gap means for API users and where Codex points instead.
· Prices as of September 28, 2026
GPT-5.5
OpenAI · Released April 23, 2026 · Previous generation
OpenAI's April 2026 flagship for coding and professional work, now a previous generation that stays in the API.
GPT-5.5 facts and comparisonsGPT-5.4
OpenAI · Released March 5, 2026 · Previous generation
The March 2026 flagship that brought GPT-5.3-Codex's coding into OpenAI's main model, now positioned as the more affordable option.
GPT-5.4 facts and comparisons
The short answer
GPT-5.5 costs exactly 2x GPT-5.4 on input, output, and cache hits, and neither charges extra to write the cache, so the example agentic coding session costs $5.00 against $2.50. GPT-5.4 left Codex for ChatGPT sign-in on August 31, 2026, and GPT-5.5 follows on October 14, 2026, while both stay in the API. OpenAI calls GPT-5.5 a new class of intelligence for coding and GPT-5.4 its more affordable model, so the choice is how much of your work needs the pricier one.
Choose GPT-5.5 if
- You want the model OpenAI launched as a new class of intelligence for coding and professional work.
- Your tasks need planning, tool use, codebase navigation, verification, and multi-step execution, which OpenAI lists as GPT-5.5's strengths.
- You run agents over large tool surfaces, where OpenAI claims precise tool use for GPT-5.5.
Choose GPT-5.4 if
- You want the lower price: $2.50 input and $15 output per million tokens, half of GPT-5.5.
- You use Codex with an API key, where GPT-5.4 is still usable after leaving ChatGPT sign-in.
- Your runs lean on compaction, which OpenAI names as a GPT-5.4 feature for longer agent runs.
- You mostly send large uncached reviews, where the example costs $0.53 against $1.05.
Side by side
Specs and prices
| Fact | GPT-5.5 | GPT-5.4 |
|---|---|---|
| Maker | OpenAI | OpenAI |
| API model id | gpt-5.5 | gpt-5.4 |
| Released | April 23, 2026 | March 5, 2026 |
| Status | Previous generation | Previous generation |
| Context window | 1.05M tokens | 1.05M tokens |
| Max output | 128K tokens | 128K tokens |
| Open weights | No | No |
| Input, per 1M tokens | $5 | $2.50 |
| Cache hit, per 1M | $0.50 | $0.25 |
| Cache write, per 1M | $5 (same as input) | $2.50 (same as input) |
| Output, per 1M tokens | $30 | $15 |
| Runs in | Codex, Cursor, OpenCode, OpenRouter, and GitHub Copilot | Codex, Cursor, OpenCode, OpenRouter, and GitHub Copilot |
Standard API rates in US dollars, as published by each maker on September 28, 2026. Batch and priority tiers, taxes, and subscription plans are not included. GPT-5.5: Prompts over 272K input tokens cost 2x for input and 1.5x for output, for the full session. GPT-5.4: Prompts over 272K input tokens cost 2x for input and 1.5x for output, for the full session.
Cost
What the same work costs
The same token counts, priced at each model’s published rates. Cached tokens are billed at each maker’s cache prices, so the session shows what caching is worth on each model.
Real sessions differ: the two models count the same code as different numbers of tokens, and reasoning settings change how much each one writes. Your own history is the real test.
| Workload | GPT-5.5 | GPT-5.4 |
|---|---|---|
| Agentic coding session, 100K input, 400K written to cache, 2M read from cache, 50K output | $5.00 | $2.50 |
| Large one-off review, 150K input with no cache hits, 10K output | $1.05 | $0.53 |
| Output-heavy generation, 30K input, 80K output | $2.55 | $1.28 |
| A month of sessions, 110 sessions: 5 a day, 22 working days | $550.00 | $275.00 |
| Where the session’s cost goes | ||
| Cache writes | $2.00 | $1.00 |
| Cache reads | $1.00 | $0.50 |
| Uncached input | $0.50 | $0.25 |
| Output | $1.50 | $0.75 |
- caching saves on the session with GPT-5.5 (64%)
- $9.00
- caching saves on the session with GPT-5.4 (64%)
- $4.50
Is GPT-5.5 exactly twice the price of GPT-5.4?
Yes. GPT-5.5 charges $5 per million input tokens, $0.50 per million cached tokens, and $30 per million output tokens. GPT-5.4 charges $2.50, $0.25, and $15. Both predate GPT-5.6, so neither adds a charge for writing the cache: written tokens cost ordinary input.
With every rate at 2x, the workloads follow. The example session costs $5.00 against $2.50, and a month of 110 sessions $550.00 against $275.00. The uncached review, $1.05 against $0.53, and the output-heavy generation, $2.55 against $1.28, land a cent off exact doubles because each total rounds to the cent.
Output is 30% of the session on both models, and cache writes are 40%. There is no write premium to earn back, so caching saves 64% of the uncached session on both, $9.00 on GPT-5.5 and $4.50 on GPT-5.4.
Both models are leaving Codex for ChatGPT sign-in
GPT-5.4 was retired from Codex for ChatGPT sign-in on August 31, 2026, and is still usable in Codex with an API key. GPT-5.5 follows on October 14, 2026. Both stay in the OpenAI API, and both are listed in Cursor, OpenCode, OpenRouter, and GitHub Copilot.
For GPT-5.4, Codex names a successor: it suggests moving to GPT-6 Sol, which the Codex docs recommend for complex coding. GPT-6 Sol follows the newer caching rules, where a cache write costs 1.25x input, so compare it on your own sessions rather than on input price alone.
If you reach these models through a ChatGPT plan, the plan meters your usage, and the dollar figures here are API-equivalent estimates. API-key users pay these rates directly, so for them the 2x gap is the real price difference.
What OpenAI says changed from GPT-5.4 to GPT-5.5
OpenAI launched GPT-5.4 in March 2026 as the flagship that brought GPT-5.3-Codex's coding into its main model, with a 1M token context window for analyzing entire codebases and compaction for longer agent runs. It now describes GPT-5.4 as a more affordable model for coding and professional work.
GPT-5.5 followed in April 2026. OpenAI says it reaches strong results with fewer reasoning tokens than earlier models at the same effort, and handles complex coding that needs planning, tool use, codebase navigation, and verification. Fewer reasoning tokens would narrow the 2x gap per task, but how much depends on your tasks.
Both have a 1.05M context window and 128K of output, and on both, prompts over 272K input tokens cost 2x for input and 1.5x for output for the full session. EveryToken reads local Codex, Cursor, and OpenCode history and prices every request at these rates, which shows whether the shorter reasoning OpenAI claims for GPT-5.5 offsets its price on your work.
Prompt caching
How OpenAI bills cached tokens
OpenAI
Prompt caching is on by default. From GPT-5.6 on you can also mark up to four explicit cache breakpoints, while GPT-5.5 and earlier cache automatically only.
From GPT-5.6 on, a cache write costs 1.25x the uncached input price and a cache hit costs 0.1x. GPT-5.5 and earlier add no charge for writing the cache: written tokens are billed as ordinary input, and a hit costs 0.1x on the models compared here.
On GPT-5.6 and later, a cached prefix stays reusable for at least 30 minutes after its last use, and caching starts at 1,024 input tokens.
Source: OpenAI: Prompt caching
Your own numbers
See what GPT-5.5 and GPT-5.4 really cost you.
everyaitoken reads your Codex, Cursor, OpenCode, and OpenRouter history on your Mac and prices every request at API rates, with what caching saved or cost. $9 once.
FAQ
Questions
Is GPT-5.4 still available?
Yes, in the OpenAI API and in Codex with an API key. It was retired from Codex for ChatGPT sign-in on August 31, 2026.
When is GPT-5.5 removed from Codex?
It goes on October 14, 2026, for anyone using ChatGPT or Codex sign-in, and stays in the API.
Do GPT-5.5 and GPT-5.4 charge for cache writes?
No. Like other OpenAI models before GPT-5.6, they bill written tokens as ordinary input and charge 0.1x input for a cache hit. The 1.25x write charge applies only from GPT-5.6 on.
Which model does Codex suggest instead of GPT-5.4?
Codex suggests moving from GPT-5.4 to GPT-6 Sol. The Codex docs also recommend GPT-6 Sol for complex coding.