Google began rolling out Gemini 4 Neon on Nov. 18, with an estimated cost of 88 cents per task at its High reasoning setting in Zapier’s launch-day AutomationBench results. Gemini 4 Carbon had an estimated cost of $2.12 per task at the same setting in that test, more than twice as much. Carbon’s Google Cloud discount runs through Nov. 30, putting its published per-token rates below Neon’s during that period.

What Changed

AI-generated summary, reviewed by an editor. More on our AI guidelines.

Business workflow tests

The test hands each agent a request inside a simulated company seeded for each run with the same starting data, including customer records, inbox threads, spreadsheets and support cases. The grader ignores the agent’s reply and checks what changed in the data. Every required condition must pass for a task to count as completed correctly.

Neon High met every scoring condition on 54.34% of AutomationBench tasks, compared with 53.88% for Carbon High in the Nov. 18 snapshot. That small score gap does not establish a capability lead. Zapier says the benchmark’s typical run-to-run variance is within 1%.

Carbon Medium led the support category at 46%. Neon’s highest overall result, 56.32% at Extra High, labeled XHigh by Zapier, met every scoring condition on just over half the tasks and cost an estimated $1.69 per attempt.

Coding tasks show the setting matters

CursorBench 4.1 provided a second check using ambiguous tasks involving multiple files drawn from real Cursor sessions. In its Nov. 18 results, Neon High scored 63.7% at an estimated $6.54 per task, and Neon Extra High scored 65.4% at $10.67. Carbon High scored 61.5% at $8.52, making Neon Extra High more expensive than Carbon High on that test.

Neon High used an average of 88,596 tokens per task, compared with Carbon High’s 114,518. Cursor estimates API costs by applying published token prices, including cache charges, to each model’s measured usage.

Cursor cautions that small score differences may not be statistically meaningful. These are estimated costs for benchmark tasks drawn from real Cursor sessions. They do not determine customer bills.

Access and model choices

The launch includes the paid Gemini API tier in Google AI Studio and the enterprise route through Gemini for business, via its API and model dropdown. Paid Google AI Pro and Ultra subscribers have access through Google Antigravity and the Gemini app.

In Antigravity, standalone Neon defaults to High. Its fourth reasoning setting appears as Extra in the model selector. Choosing Orchestrator with Neon pairs a Neon planner with Hydrogen carrying out implementation work, so that option uses two models during a task.

Know someone who'd find this useful? ✉️ Email it to a friend in one click, or they can subscribe free here.

Google positions Hydrogen for fast everyday execution and Carbon for its hardest, longer projects.

Token prices change the comparison

On Nov. 18, Google Cloud’s Global standard rates for Neon were $3 per million input tokens and $15 per million output tokens, in U.S. dollars. In common text tasks, input tokens are pieces of text sent to the model. Output tokens cover the text it generates, including reasoning as well as the response. Task cost depends on how many tokens it uses.

Carbon’s introductory Global standard rates are $2 per million input tokens and $10 per million output tokens through Nov. 30. Its introductory discount is delivered through credits against eligible spending. The Cloud schedule does not establish which rates Zapier or Cursor applied.

Under the published Google Cloud schedule, Carbon’s Global standard rates rise above Neon’s on Dec. 1, to $4 per million input tokens and $20 per million output tokens.

Frequently Asked Questions

How much does Gemini 4 Neon cost?

On Nov. 18, Google Cloud’s Global standard rates were $3 per million input tokens and $15 per million output tokens. Output charges include reasoning tokens.

How did Neon compare with Carbon on Zapier’s test?

Neon High met every scoring condition on 54.34% of tasks at an estimated 88 cents per task. Carbon High scored 53.88% and averaged an estimated $2.12 on the same test.

What does more reasoning cost on the coding test?

In CursorBench 4.1’s Nov. 18 results, Neon Extra High cost an estimated $10.67 per task, compared with $6.54 for Neon High. The two settings scored 65.4% and 63.7%, respectively.

Where is Google rolling out Neon?

The Nov. 18 rollout covers the paid Gemini API tier in Google AI Studio and Gemini for business. Paid Google AI Pro and Ultra subscribers can access Neon in Google Antigravity and the Gemini app.

When do Carbon’s introductory rates end?

Google Cloud’s Global introductory rates of $2 per million input tokens and $10 per million output tokens run through Nov. 30. On Dec. 1, the scheduled rates are $4 and $20, respectively.

AI-generated summary, reviewed by an editor. More on our AI guidelines.

Google Chases OpenAI Into the Decision-Model Market
Google released the Gemini Decisions API on Wednesday, an endpoint that answers closed-set questions instead of writing text. Its new model, Gemini Decision 1, can take up to 1,000 questions about one
OpenAI's GPT-6.2 Squeezes More Work Out of Each Dollar
Ethan Mollick, the Wharton professor who tests most models before release, asked GPT-6.2 Astra to build a space elevator ride from scratch in Blender and Unity. In his account of the test, he said it
Which AI Coding App Should You Use?
  The Decision Brief Which AI coding app should you use, and which can you rely on without tying yourself to one AI company? This guide profiles 19 apps, paid and free, from Claude Code,
AI News

San Francisco

Editor-in-Chief and founder of Implicator.ai. Former ARD correspondent and senior broadcast journalist with 10+ years covering tech. Writes daily briefings on policy and market developments. Based in San Francisco. E-mail: editor@implicator.ai