Wink Pings

Gemini 4 Neon Release: 1 Million Token Context Window, No.1 on Coding Agent Leaderboard

Google DeepMind has launched Gemini 4 Neon, which supports a 1 million token context window, with input/output pricing of $3/$15 per million tokens. It scores 61 on the AA Intelligence Index and takes the top rank on the AA Coding Agent Index.

Google DeepMind has officially released Gemini 4 Neon. As reported by MarkTechPost, this is a Gemini 4 model equipped with a 1 million token context window, with input priced at $3 and output at $15 calculated per million tokens. It scores 61 on the AA Intelligence Index and ranks first on AA's Coding Agent Index.

Let’s first explain two key terms. Context length can be interpreted as the maximum volume of content the model can process in a single session. A 1 million token context window means large codebases, full long documents, or multi-turn tool call histories can be imported into the model all at once. Token is the standard billing unit for large language models, with input and output counted separately. Developers building applications need to calculate both of these two costs in their budget.

The AA Intelligence Index and Coding Agent Index are two separate evaluation leaderboards. The former assesses the overall general intelligent performance of models, while the latter focuses more on model capabilities in coding agent scenarios. Gemini 4 Neon taking the top spot on the latter ranking shows Google is emphasizing far more than just chat functionality for this model: it is also designed to write code, call and adjust tools, and complete full end-to-end task chains.

For developers, this release sends two direct signals. First, the 1 million token context window is now available in the Gemini 4 series, meaning long context is no longer just an experimental parameter limited to lab environments. Second, the official pricing is now public: whether the $3/$15 per million tokens rate is acceptable depends on your specific workload. Long document summarization and coding agent use cases have completely different consumption patterns, so you cannot evaluate cost effectiveness only by the unit price.

Leaderboard scores have reference value, but do not treat them as a substitute for the full real-world user experience. When coding agents are actually deployed for production use, factors including latency, tool call stability, error recovery ability and permission control will all affect the final performance. Whether Gemini 4 Neon can become part of your daily toolchain still depends on how it performs in these granular practical details.

Source: MarkTechPost, by Asif Razzaq.
Original article link: https://www.marktechpost.com/2026/11/18/google-gemini-4-neon-release/

发布时间: 2026-11-19 14:52