Google launches Gemini 4 Neon model with million-token context window
仅事实

Google launches Gemini 4 Neon model with million-token context window

摘要

Google introduced Gemini 4 Neon, its fourth Gemini 4 variant in under two months, offering a one-million-token context window and competitive benchmark scores.

Google released Gemini 4 Neon on Nov. 18, marking the fourth Gemini 4 model launched within a two-month period. The model is positioned between the deeper-reasoning Carbon and the faster-execution Hydrogen, according to Google’s product positioning.

Neon is available through Google’s API, AI Studio, and enterprise products, and can be accessed by Pro and Ultra subscribers via the Gemini app and Antigravity. It supports a context window of one million tokens.

In Google’s model-card comparison, Neon outperformed Anthropic’s Claude Opus 5.5 and OpenAI’s GPT-6.2 Sol in 16 of 19 benchmark categories, beating both rivals in 13 of the 16 categories where all three models were evaluated. Notable scores include 95.2 on the GraphWalks test, compared with 66.8 for Opus and 80.1 for Sol. Opus led on Terminal-Bench 4.1 and FrontierCode 2.0, while Sol performed best on Chartography.

Google notes that Neon, adapted from Carbon, can still produce hallucinations. The company highlights strong results in long-context, automation, and cybersecurity tests, but leaves the model’s performance on real-world documents and workflows to be evaluated by developers.

Pricing for Neon’s API is set at $3 per million input tokens and $15 per million output tokens, 25% lower than Carbon’s announced standard rates of $4 and $20. Carbon’s introductory discounted rates of $2 for input and $10 for output expire on Nov. 30, making Neon’s current rates about 50% higher than the discounted Carbon pricing.

The release follows a period in 2026 when Google faced criticism for slow access to its most capable models. Earlier reports indicated missed release deadlines and limited rollouts of previous models such as Argon. By extending Neon to Pro subscribers, Google broadens access beyond the Ultra tier required for Carbon’s consumer launch, offering developers an additional option with a different balance of reasoning depth, speed, and cost.

Independent measurements by Artificial Analysis on Nov. 19 gave Neon an Intelligence Index score of 61 at its highest reasoning setting and 60 at a high setting, with task completion times ranging from 14.8 to 21.5 minutes. Developers are advised to assess whether the increased reasoning capability justifies longer processing times and to compare cost per successful task, especially given Carbon’s temporary discount.

继续阅读——免费

创建免费账户以关注新闻。无需银行卡。