500xEngineer.

All postsSHORT2 min read

Gemini 4 Neon Costs More Per Token Than Carbon This Month

Gemini 4 Neon's token prices, Carbon's December price change, Google's task-cost claim and Artificial Analysis's findings at xhigh.

Gemini 4 Neon costs more per token than Google's flagship Carbon this month. So why pick it? Google reports a lower cost per task at Neon's default High setting. Carbon's current rates also have an end date: November 30, 2026.

Video not loading? Watch the Short on YouTube

What Neon is

Google released Gemini 4 Neon on November 18, 2026, for complex software engineering, knowledge work and science. Google positions it between Carbon for the hardest, long-horizon projects and Hydrogen for fast everyday execution.

You can access Neon through the Gemini API in Google AI Studio. Google also announced it for Gemini for business, paid Google AI Pro and Ultra subscribers in Antigravity, and Pro and Ultra subscribers in the Gemini app.

The prices and Carbon's December change

These are standard API rates per 1 million tokens. Output pricing includes thinking tokens.

Model and period Input Output
Neon $3 $15
Carbon through November 30, 2026 $2 $10
Carbon starting December 1, 2026 $4 $20

Neon's cached input is $0.15 per million tokens, and its context window is 1 million tokens. There's no free tier for Neon in the Gemini API.

The calendar changes the comparison. Neon has higher input and output rates than Carbon in November. Carbon's rates double on December 1, putting both above Neon's listed rates.

What Google claims

Google says Neon ranks #1 on the Artificial Analysis Coding Agent Index with a score of 58.6. It also says Neon costs 32% less per task than Carbon at its default High thinking level. Google doesn't say which Carbon price, November's or December's, that comparison uses.

Google reports that Neon ranks #1 on most Surge AI benchmarks across finance, accounting and spreadsheet modeling. Those are Google's reported rankings. I'd use them as a reason to test a workload, then check the actual result.

What Artificial Analysis measured

Artificial Analysis calls Neon at its highest setting, xhigh, very verbose and notably slow. On its Intelligence Index, Neon generated 283M output tokens against a 65.1M median. That's over 4 times the median output.

At xhigh, tasks averaged 21.5 minutes against a 12.1-minute median. Artificial Analysis reports an average evaluation cost of $6.44 per task at that setting.

Keep the settings attached to those numbers. Google's 32% lower task-cost claim uses default High. The strongest verbosity and speed warnings above are Artificial Analysis's xhigh measurements.

Which setting I'd use and how I'd watch spend

Neon has four thinking levels, from Low to Extra High. Artificial Analysis measured 28.3M output tokens at Low, 62.9M at Medium, 123M at High and 283M at xhigh on its Intelligence Index. Average task times were 3.7, 8.3, 14.8 and 21.5 minutes respectively.

My take: I'd start at default High, then compare a lower setting on the same task. I'd check the answer before deciding whether the extra thinking was useful. I'd watch billed output, including thinking tokens, and consider cached input for repeated material at the listed $0.15 per million tokens.

Here's how you get paid

My opinion: I'd sell bookkeepers or small firms a month-end statement check. I'd ask Neon to flag odd charges across statements, receipts and spreadsheets. A person would check every flag before anything went to the client.

Google's reported finance, accounting and spreadsheet results make that an offer I'd test. I'd keep human review in the deliverable. No income guaranteed.

Daily email

Get each breakdown in your inbox.

One email each evening, right after the Short goes live: the AI tool it covers, the full write-up with sources, and how to get paid for it. Free, and you can unsubscribe anytime.

Read past emails on Substack

Sources

All postsWatch on YouTube