Home / Models / OpenAI / GPT-5.5

GPT-5.5 pricing and routing profile

GPT-5.5 from OpenAI lists at $5.00 per million input tokens and $30.00 per million output tokens, with a 1,050,000-token context window. Prices from OpenRouter's catalogue on 2026-09-28.

Input, per 1M tokens$5.00
Output, per 1M tokens$30.00
Cached input, per 1M$0.50
Context window1.05M
Confidence signalNot exposed

Tool callingStructured outputReasoning control

Cost per 1,000 requests

Request shapeList cost
Classification or routing (600 tokens in, 10 out)$3.30
Chat reply (2,000 in, 400 out)$22.00
Retrieval-augmented answer (8,000 in, 500 out)$55.00

Before prompt caching. Try your own shape in the cost calculator.

Can Gimbix route to it first?

Not as a first-answer model through OpenRouter: no host returns token log probabilities for GPT-5.5. It can still be the model Gimbix falls back to when a cheaper model is unsure.

Who serves GPT-5.5

HostInput / output per 1MLog probabilitiesUptime, last 30 min
OpenAI$2.50 / $15.00no—
Azure$5.00 / $30.00no100.0%
OpenAI$5.00 / $30.00no100.0%
Azure$5.50 / $33.00no—
Azure$5.50 / $33.00no—
Amazon Bedrock$5.50 / $33.00no—
OpenAI$12.50 / $75.00no—

A host that lists log probabilities may still omit them; Gimbix checks each one before relying on it.

Cheaper models that can answer first

Models priced below GPT-5.5 that expose a confidence signal. Whether they answer your traffic as well is what a Gimbix pilot measures.

ModelInput / output per 1MCheaper on a chat reply
Qwen3.8 Max Prime$4.00 / $12.0042%
Ember-1$3.00 / $15.0045%
Kimi K3$3.00 / $15.0045%
GLM 5.3 Prime$2.80 / $8.8059%
GPT Audio$2.50 / $10.0059%
GPT-4o (2024-11-20)$2.50 / $10.0059%

Questions

How much does GPT-5.5 cost?

GPT-5.5 lists at $5.00 per million input tokens and $30.00 per million output tokens on OpenRouter as of 2026-09-28. A typical chat reply of 2,000 input and 400 output tokens costs about $22.00 per 1,000 requests.

What is the context window of GPT-5.5?

1,050,000 tokens.

Can a cheaper model replace GPT-5.5?

Sometimes. Gimbix proves it on your own traffic before routing: the cheaper model must match GPT-5.5 on one half of your requests and make at most three answers worse on the other.