Models

Gemini 4 Neon cuts task cost while intelligence gain stays modest

Artificial Analysis found Gemini 4 Neon scores 61 on its Intelligence Index, generates 110 tokens per second against Carbon’s 57 and costs $6.44 per task, with mixed results at higher effort.

Gemini 4 Neon Delivers Small Intelligence Gain at Lower Cost Per Task
Gemini 4 Neon Delivers Small Intelligence Gain at Lower Cost Per Taskpixifant / pixabay
By Rebecca Stone6 min read

Why it matters

  • Gemini 4 Neon became generally available on November 18, 2026.
  • Neon (xhigh) scored 61 on the Intelligence Index, one point above Carbon (high).
  • Neon cost $6.44 per Intelligence Index task and averaged about 21 minutes, versus $7.76 and about 36 minutes for Carbon at standard pricing.
  • Neon generated about 110 tokens per second against Carbon’s 57, despite using roughly 15% more output tokens per task.
  • On GDP.pdf, Neon (high) posted a 33.4% all-pass score, while Neon (xhigh) posted 28.0%.

Artificial Analysis measured Google’s Gemini 4 Neon (xhigh) at 61 on its Intelligence Index on November 18, 2026, one point above Gemini 4 Carbon (high). The report put Neon at $6.44 per Intelligence Index task and about 21 minutes per task, against $7.76 and about 36 minutes for Carbon at standard, non-discounted pricing.

The ranking frames the release as an efficiency contest rather than a clear intelligence lead. Artificial Analysis placed Neon third overall, behind GPT-6.2 Astra (max) and Claude Fable 5.2 (max), with the cost and speed gains doing more to differentiate it than the one-point benchmark advantage.

What did Google release?

Google made Gemini 4 Neon generally available on November 18, 2026, according to its release notes. Google positioned it for complex software engineering, knowledge work and science workflows.

Google’s release notes list four thinking levels, a unified 1M-token context window and long decode support of up to 1M tokens.

Google’s release notes also describe Neon as a proprietary reasoning model. It accepts text, image, speech and video input, produces text, and uses high as its default effort setting.

At launch, Neon reached Google’s API and the Gemini Enterprise Agent Platform, according to Google’s release notes. Google also offered it in Antigravity and the Gemini app to Pro and Ultra subscribers.

Why did Neon complete tasks faster?

Artificial Analysis found Neon used about 15% more output tokens per task than Carbon. Its lower cost came from pricing, not reduced token use.

Neon produced output at about 110 tokens per second, against about 57 for Carbon, according to Artificial Analysis. The report said that speed accounted for the shorter completion time.

At high effort, Artificial Analysis recorded a score of 60 and 14.8 minutes per Intelligence Index task for Neon. The setting matched Carbon (high) on intelligence while costing less than half as much per task under standard pricing, according to Artificial Analysis.

Effort level changes the result by workload, not in a straight line. Neon (xhigh) scored 61 on the Intelligence Index, while Neon (high) scored 60. On GDP.pdf, the order reversed: high led xhigh. Artificial Analysis reported that performance still varies by effort level and workload, so the highest setting does not dominate every test.

How did Neon perform as a coding agent?

Artificial Analysis used Antigravity CLI to measure coding agents. Its detailed results were:

  • Gemini 4 Neon (xhigh): 59 on the Coding Agent Index, with an average cost of $6.86 per coding-agent task.
  • Gemini 4 Carbon (high): 55 on the Coding Agent Index.
  • Claude Fable 5.2 (xhigh): 58 on the Coding Agent Index, with an average cost of $15.47 per task.
  • Claude Opus 5.5 (max): 58 on the Coding Agent Index, with an average cost of $12.61 per task.

Neon’s score was one point above each Claude result, while its task cost was lower than both, according to Artificial Analysis. The combination supports Neon’s operational advantage in coding-agent use without giving it a large intelligence lead.

Why did higher effort underperform on documents?

On GDP.pdf, a professional-document reasoning evaluation, Artificial Analysis measured a 33.4% all-pass score for Neon (high), versus 28.4% for Carbon (high) and 28.0% for Neon (xhigh).

Artificial Analysis defines all-pass as the share of 500 attempts per model, across 100 tasks in 10 professional domains, in which every criterion passes.

The xhigh result came below the high setting on GDP.pdf, according to Artificial Analysis. The document test cautions against choosing an effort setting from benchmark rank alone.

What did the knowledge-work evaluations show?

On AA-Briefcase, Artificial Analysis measured 1700 Elo overall for Neon (xhigh) on a comparative scale for agentic knowledge work. The model also scored 1771 Elo for Presentation Quality.

That presentation score stood 45 points above the previous leader, Claude Sonnet 5.5 (max), which had 1726, according to Artificial Analysis.

The separate Multimodal Knowledge Agent results put Neon’s Climb WSR at 56%, two points above Carbon and just below Claude Fable 5.2 (max) at 57%.

Artificial Analysis defines Climb WSR as the mean weighted satisfaction rate on harder criteria in Google DeepMind’s evaluation of agent work on visually dense project files. MKA is a standalone evaluation, not an Intelligence Index component.

The evaluations use different units because they measure different workloads. GDP.pdf reports an all-pass percentage, AA-Briefcase reports comparative Elo, and MKA reports Climb WSR. The figures should not be combined into a single score: Artificial Analysis identifies MKA as a standalone evaluation, not an Intelligence Index component.

What will buyers pay?

Google charges $3 per million Neon input tokens and $15 per million output tokens on its paid API tier. It prices cached input at $0.15 per million tokens after a 95% discount.

Through November 30, 2026, Carbon’s discounted rate is $2 per million input tokens and $10 per million output tokens. Artificial Analysis said Neon sits 50% above that temporary rate and 25% below Carbon’s standard $4/$20 input/output rate.

The measured task costs and list prices serve different purposes. The $6.44 Intelligence Index task and $6.86 coding-agent task are workload averages from Artificial Analysis.

The $3/$15 Neon rates and Carbon’s $2/$10 promotion set per-token charges. Artificial Analysis treats Carbon’s promotion as a price change, so the standard-price task comparison and the current token-price comparison can point in different directions.

How far does the benchmark generalize?

Artificial Analysis cautioned that its Intelligence Index methodology “may not apply directly to every use case.” The index combines nine evaluations across knowledge work, coding, science and general knowledge, with each category weighted at 25%.

The suite is primarily English-language and text-based. Image-capable models receive images in some evaluations, according to Artificial Analysis.

Artificial Analysis evaluates visual reasoning, speech input and multilingual performance separately. The equal 25% weighting gives knowledge work, coding, science and general knowledge equal influence, even though the index contains nine evaluations.

Neon’s results describe a modest composite intelligence advance with a larger documented advantage in operational speed and measured task cost under standard pricing. Performance still varies by effort level and workload, while Carbon’s temporary discount changes the immediate token-price comparison.

For enterprises assessing the model, the immediate test is which workloads it can handle, which effort level those workloads require and what each completed task costs. Carbon’s temporary rate expires November 30, 2026, giving buyers a dated checkpoint for comparing Neon’s standard pricing with Carbon’s undiscounted rates.

Source: unite.ai

Share this article:

More from Rebecca Stone

Rebecca Stone

Show full bio

Correspondent covering companies, deals and strategy in artificial intelligence and machine learning at AI In Context.

802 articles

Related articles

  1. Gemini 4 Neon runs at 110 tokens per second, undercuts Claude
  2. Google Ships Gemini 4 Neon, a Cheaper Near-Carbon Model at $3 Per Million Tokens
  3. Google Opens Gemini 4 Neon to All Developers at $3/$15 per Million Tokens
  4. Google Launches Gemini 4 Hydrogen, Beating Argon on Intelligence Index
  5. Gemini 4 Carbon tops AI benchmarks with guarded security rollout

« Previous articleNext article »