Google 發佈 Gemini 4 Neon,在多項基準測試中超越 Claude Opus 5.5
Google's New Gemini 4 Neon Outperforms Claude Opus 5.5 on Several Benchmarks
Google 發佈 Gemini 4 Neon,並稱其在編碼及智能體等多項基準測試中超越 Claude Opus 5.5。
文章並列 Gemini 4 Neon 的基準領先項目、Claude Opus 5.5 保有優勢的測試及 API 價格,可對照其能力表現與使用成本。
Google has released Gemini 4 Neon, its latest AI model, claiming performance that surpasses Anthropic's Claude Opus 5.5 on several benchmarks, including coding and agentic tasks.
The model ranks first on the Artificial Analysis Coding Agent Index, which combines results from multiple coding benchmarks and measures how reliably agents can implement code changes, complete technical tasks, and understand existing codebases.

Google said Neon outperforms Claude Opus 5.5 on selected evaluations, including CursorBench 4.1.
However, Opus 5.5 retains a narrow lead on Terminal-Bench 4.1 and FrontierCode 2.0.
Google has priced Gemini 4 Neon at $3 per million input tokens and $15 per million output tokens through its Gemini API. Cached input tokens cost $0.15 per million, representing a 95% discount against the standard input rate.

Beyond its API pricing, Gemini 4 Neon also shows a cost advantage on complex reasoning tasks, while performing better than competing frontier models.

The model also recorded leading scores in several specialised evaluations. It ranked first on the Legal Research Bench and Tax Agent Bench.
Gemini 4 Neon is rolling out across the Gemini API through Google AI Studio, Gemini for Business, and Google Antigravity. Google AI Pro and Ultra subscribers can access it through the Gemini app, while paid subscribers can also use it in Antigravity.
Google's Gemini 4 model family continues its element-based naming convention, with Carbon, Hydrogen, Argon, and now, Neon.
The company has been rolling out the Gemini 4 model family in stages since September, beginning with Gemini 4 Argon, its flagship frontier model.
This was followed by Gemini 4 Carbon, introduced in October and made available to paid subscribers in November, and Gemini 4 Hydrogen, launched recently for faster, everyday tasks and subsequently integrated into Google Search's AI Mode.
來源:Analytics India Magazine · analyticsindiamag.com