Artificial intelligence has a new competitive focus in the fields of complex reasoning and multi-step task automation. On November 18, Google officially launched a new AI model, Gemini4Neon, focusing on complex programming, scientific research, and enterprise workflows, and aiming to find a better balance among reasoning capability, response speed, and operating cost. The model is now being made available to developers, enterprises, and some paid subscribers.
Core capabilities: A million-token ultra-long context and multi-agent collaboration
According to Google, Gemini4Neon has a context window of up to 1 million tokens. It can comprehensively process various forms of data such as spreadsheets, PDFs, charts, audio, and video, and consolidate scattered information into structured reports and other outputs.

In software development and automated execution, Neon demonstrates strong multi-step processing capabilities:
Code and engineering tasks: It supports cross-file error troubleshooting and code refactoring, making it suitable for handling complex issues in large codebases.
Multi-agent orchestration: In Google's Antigravity development environment, Neon can act as a "dispatcher," coordinating Hydrogen sub-agents to process different tasks in parallel and continuously synchronizing progress during long-running operations.
Performance: No. 1 on the coding agent index, surpassing Claude Opus5.5
In benchmark tests and internal evaluations, Google released several sets of targeted data:
Authoritative leaderboard results: At the highest xhigh reasoning level, Neon ranked third with 61 points on the Artificial Analysis Intelligence Index; on the same organization's Coding Agent Index, it scored 58.6, surpassing Claude Opus5.5 to take first place.
Efficiency and quality comparison: Early internal evaluations show that Neon's overall task quality matches or exceeds that of the larger Gemini4Carbon, while completing tasks three times faster. At the default high thinking level, the cost per task in related tests was 32% lower than that of the same-series Carbon model.
Google also noted that the above rankings and cost comparisons correspond to specific evaluation conditions and do not represent performance in all real-world business scenarios.
Pricing and access: Four reasoning levels available, gradually rolling out across platforms
For API call pricing, Gemini4Neon's specific fees are as follows:
Input cost: $3 per 1 million tokens (approximately RMB 20 at current exchange rates)
Output cost: $15 per 1 million tokens (approximately RMB 100.2 at current exchange rates)
Cached input cost: $0.15 per 1 million tokens (approximately RMB 1 at current exchange rates)
The model offers four reasoning levels: Low, medium, high, and xhigh, allowing users to adjust thinking depth and resource consumption as needed. For individuals and subscribers, after subscribing to Google AI Pro or Ultra, they can choose to use Gemini4Neon in the Gemini app.
Join Now