Google priced Gemini 4 Neon at $3 per million input tokens and $15 per million output tokens in its launch post on November 18, 2026. At the Bank of Canada rate for November 19, 2026, that is $4.30 CAD per million input tokens and $21.48 CAD per million output tokens.
Direct answer: Neon sits between Carbon and Hydrogen in the Gemini 4 family. It scores 61 on the Artificial Analysis Intelligence Index, rank 3 overall at its highest thinking level. Google states that Neon at its default high setting costs 32% less per finished task than Carbon. The token price is higher than Carbon’s launch price. The cost per task is lower. The second figure is the 1 your bill follows.
| Measure | US price | Canadian price |
|---|---|---|
| Input tokens, per million | $3 | $4.30 CAD |
| Cached input tokens, per million | $0.15 | $0.21 CAD |
| Output tokens, per million | $15 | $21.48 CAD |
The Canadian figures convert Google’s published US prices at US$1 = CAD 1.4317, the Bank of Canada daily rate for November 19, 2026.
What does Neon claim on the test scores?
Google reports rank 1 on the Artificial Analysis Coding Agent Index, with a score of 58.6. It reports 65.2% on Terminal-Bench 4.1. It reports rank 2 on steerability, which measures how cleanly a model takes a correction in the middle of a task.
The context window is 1 million tokens. The model has 4 thinking levels, and high is the default. Each level trades token spend against depth of reasoning on the same job.
Treat every score in this section as vendor reported. Google published the table, and independent runs will settle over the next weeks. No hands-on test of Neon was run for this column.
Why does cost per task beat cost per token here?
A token price counts words. A task price counts finished work, with the retries inside it. A model that answers in 1 pass at a higher token price can cost less than a model that needs 3 passes at a lower 1.
That is the same trap we priced in Gemini 4 Hydrogen costs $2.14 CAD. Speed is the real change. Hydrogen won on speed for daily work. Neon claims the middle seat. It aims at complex coding, analytical reasoning and production agent work, at a speed Carbon does not match.
The comparison set keeps moving. GPT-6.2 Sol costs $2.87 CAD. Regular Chat gets neither model set the rival input price this week. Qwen4-Max output costs $6.42 CAD. Route the easy steps, keep client data home showed the routing answer. A Canadian buyer now holds 3 live price cards for the same class of work, and each card counts a different unit.
Where can a Canadian team get Neon today?
Google lists the Gemini API in Google AI Studio and Android Studio for developers. Gemini for business carries it through the API and the model dropdown. Paid Google AI Pro and Ultra subscribers can select it in the Gemini app and in Google Antigravity.
Neon for cyber rolls out to trusted defenders in the coming days under the Fairwind Program. That path runs through a Google Cloud account team, so a Canadian security team starts with its account rep, not with a signup page.
What test should you run before you switch a workload?
Run 1 week of real work on Neon and on your current model, side by side. Pick 20 finished tasks your team already repeats. Coding fixes, analysis runs, document builds. Count 3 figures per task: total tokens, number of retries, and whether the output shipped without a human rebuild.
Divide the week’s spend by the tasks that shipped. That quotient is your cost per finished task, in CAD, on your own work. It answers the only question the vendor table cannot answer for you.
The acceptance habit in Count accepted comments, not posted ones, before you renew your AI review tool is the same discipline in a different tool. Count what your team accepted. Ignore what the tool produced.
Recap: Neon costs $4.30 CAD per million input tokens and $21.48 CAD per million output tokens. It ranks 3 overall on the index Google cites, and Google claims a 32% lower cost per task than Carbon at the high setting. The claim is testable in 1 week on your own tasks.
Before you move any workload, run the 20 task test and put the CAD cost per shipped task in writing. If Neon wins on your work, switch the workload. If it wins only on the vendor table, keep your current model.