Google’s Gemini 4 Neon Challenges Claude Opus 5.5 With Similar Coding Power at 25% Lower Prices
The AI race is entering a new phase. The winner may not be the company with the smartest model, but the company that delivers the best results at the lowest total cost.
Gemini 4 Neon Beats GPT-6.2 Sol on Several Major AI Benchmarks
Google has introduced Gemini 4 Neon, a mid-range AI model designed to handle demanding professional work without charging flagship prices. Its most interesting achievement is not simply its benchmark performance. It is the combination of speed, coding ability and pricing.
At $3 per million input tokens and $15 per million output tokens, Neon costs 25% less than Anthropic's Claude Opus 5.5 at standard rates. Yet Google reports that Neon comes within roughly one percentage point of Opus on two engineering benchmarks.
For businesses spending money on AI every day, that difference deserves attention.
Google Introduces a New Middle Ground in AI
Gemini 4 Neon is available through Google's paid Gemini API, Google AI Studio and Android Studio. It is also available through the Gemini app and Google Antigravity for eligible Pro and Ultra subscribers.
Google positions Neon between two existing models.
-
Hydrogen is designed for smaller, faster, high-volume tasks.
-
Neon targets demanding work that businesses perform regularly.
-
Carbon is intended for the most difficult, extended projects.
This creates a more flexible product range. Customers do not need to pay for the most powerful model every time they ask AI to perform a task.
That is a smart commercial strategy.
Performance Is No Longer Just About Model Size
Google says Neon matched or exceeded Gemini 4 Carbon on overall quality in its early evaluations, while running three times faster and costing 60% as much.
However, Google did not publish absolute completion times for an identical Neon and Carbon workload. The speed and cost claims should therefore be treated as reported customer evaluation results, not universal guarantees.
The independent benchmark picture is also interesting.
|
Model |
Intelligence Index |
Vals Index |
|---|---|---|
|
Gemini 4 Neon |
61 |
70.9% |
|
Claude Opus 5.5 |
59 |
70.14% |
|
GPT-6.2 Sol |
57 |
66.06% |
|
Claude Sonnet 5.5 |
55 |
70.85% |
|
Gemini 4 Hydrogen |
54 |
68.49% |
|
Gemini 4 Carbon |
60 |
70.45% |
Neon ranks first among these models on the listed Artificial Analysis Intelligence Index results.
It also scores slightly above Opus and Sonnet on Vals, an evaluation covering finance, software, legal and tax work.
But benchmark leadership does not mean universal superiority. Different models perform differently depending on the task.
The Real Advantage Is the Cost of Finishing Work
Neon's standard pricing is $3 per million input tokens and $15 per million output tokens.
Compare that with Claude Opus 5.5 at $4 and $20.
25% lower standard token prices than Opus 5.5
However, Neon is not the cheapest competitor.
GPT-6.2 Sol and Claude Sonnet 5.5 both cost $2 per million input tokens and $10 per million output tokens at standard short-context rates.
That means Neon carries a 50% premium over those two models.
For a business, the important question is whether Neon completes work with fewer tokens, less human intervention or fewer repeated attempts.
If it does, the higher price per token may still produce a lower final bill.
At the same input-output mix, Neon would need to use approximately one-third fewer tokens than Sol or Sonnet to offset its 50% price premium.
That is the calculation procurement teams should focus on.
Speed Changes the Developer Experience
According to launch-day benchmark tracking, Neon generates approximately 110 tokens per second at its highest reasoning setting.
Gemini 4 Carbon reaches approximately 59 tokens per second at its highest listed setting.
That makes Neon roughly 1.9 times faster in output generation based on those readings.
But there is an important limitation.
Neon takes approximately 57 seconds to produce its first token at the highest reasoning setting.
For software developers, generating text quickly is only part of the process. An AI coding agent must also inspect files, understand the project, run tests and correct mistakes.
A model that generates code quickly but requires extensive revisions may not actually save time.
Google Antigravity Shows the Value of Software Around AI
One of the strongest findings in the announcement concerns the software environment.
On the SWE-sweep evaluation, Google reports:
-
12.8% using the mini-SWE agent software.
-
39.0% using Google Antigravity.
The underlying Neon model remained the same.
That is approximately three times the score.
Why does this matter?
Because AI performance does not come from the model alone. The surrounding software determines how effectively the model can use tools, divide work and complete projects.
Antigravity allows Neon to distribute a project among multiple Hydrogen agents working in parallel.
This is a significant architectural advantage.
Google is not simply selling access to an AI model. It is building an environment in which multiple models can cooperate to complete complex engineering tasks.
For businesses, that may be more valuable than another small improvement on a benchmark leaderboard.
The Professional Market Is Becoming More Competitive
Neon also produced encouraging results in specialized tasks.
Google reports 70.9% on the Vals Index, placing it slightly above Claude Opus 5.5 at 70.14%.
Neon also ranked first on the evaluation's legal research and tax agent tasks.
On Zapier's AutomationBench, Neon scored 56.3%, compared with 42.5% for Opus and 41.3% for Sol.
However, the results are not consistent across every category.
Opus scored 65.8% against Neon's 65.2% on Terminal-Bench 4.1. Sol scored 75.1% against Neon's 73.5% on Chartography.
The message is clear. No single leaderboard tells businesses which model is best for every department.
Google Is Building a Platform, Not Just a Product
The bigger strategic opportunity is Google's ability to connect models, development tools, cloud infrastructure and consumer subscriptions.
Neon gives Google a competitive middle tier.
Hydrogen handles volume. Neon handles recurring professional work. Carbon targets more demanding projects.
This structure creates opportunities to keep customers inside Google's ecosystem.
A developer can experiment in AI Studio, build applications through development tools and deploy workflows using Google's broader infrastructure.
That creates a stronger business relationship than selling individual API calls alone.
It also increases the importance of switching costs. Once a company builds workflows around Google's tools, changing providers may require more than replacing one model name.
Anthropic and OpenAI Face a Different Pricing Challenge
Anthropic has a powerful flagship in Opus 5.5, while OpenAI offers competitive pricing through GPT-6.2 Sol.
Google's challenge is to convince customers that Neon provides a better balance of quality and operating cost.
Its biggest advantage is not that it is always cheaper. It is that Google offers a range of models and an integrated development environment.
Anthropic and OpenAI can respond through better models, lower prices, faster inference or improved developer tools.
This competition should benefit customers.
However, benchmark differences must be evaluated carefully because reasoning settings, workloads and testing environments vary.
The Biggest Weakness in the Story Is Cost Per Completed Task
From a competition and market-regulation perspective, the most important question is whether Google's performance claims translate into measurable economic benefits for customers.
The weakest part of the announcement is the absence of a complete, independently comparable cost-per-completed-task analysis across Neon, Sol and Sonnet.
Google reports a 32% reduction in cost per task against Carbon at Neon's default high setting. At the highest reasoning setting, that advantage narrows to approximately 17%.
These are useful figures, but buyers still need comparable results across their own workloads.
A model with lower token prices can become more expensive if it needs more attempts, more reasoning or more human corrections.
Companies should publish clearer information about benchmark configurations, completion times and total operating costs.
Otherwise, customers risk buying a convincing leaderboard position rather than a genuinely better business tool.
What Small Businesses Should Do
Small and medium-sized businesses often face a difficult choice. They want advanced AI capabilities but cannot afford expensive experimentation.
Neon creates another option, but businesses should avoid moving every workflow immediately.
A practical strategy is to:
-
Test Neon against Sol and Sonnet using identical business tasks.
-
Measure the total cost of producing finished work.
-
Compare the number of corrections required.
-
Use Hydrogen for simple, repetitive tasks.
-
Reserve Neon for complex coding, analysis and automation.
-
Use Carbon only when its additional capabilities justify the premium.
For a small software company, even a modest reduction in development time can make a meaningful difference.
The goal should not be to use the most powerful AI model. It should be to complete valuable work at the lowest reliable cost.
The Next AI Winner May Be the Company That Delivers More for Every Dollar
Gemini 4 Neon demonstrates how quickly the AI market is changing.
Google is combining competitive intelligence, faster output, professional capabilities and a layered model strategy.
But its most important contribution may be its broader platform approach.
The future of AI will not be decided by benchmark scores alone. It will depend on how much useful work businesses can complete, how much that work costs and how reliably the software delivers the result.
For customers, this is good news.
AI is becoming less like an expensive experiment and more like practical business infrastructure.
Gemini 4 Neon Beats GPT-6.2 Sol on Several Major AI Benchmarks
Gemini 4 Neon, Google AI model, Gemini 4 pricing, AI coding model, Claude Opus 5.5, GPT-6.2 Sol, AI token pricing, Google Antigravity, AI coding tools, enterprise AI, AI automation, model benchmarks, AI productivity, AI development platform