Google releases Gemini 4 Neon: third on the Artificial Analysis Intelligence Index at 61, priced at $3 per million input tokens
Synopsis
Google introduced Gemini 4 Neon, a frontier model positioned for fast end-to-end task completion across complex software engineering, knowledge work and science workflows. At its xhigh thinking level it scores 61 on the Artificial Analysis Intelligence Index, third overall, and ranks second on the Vals Index on the Pareto frontier; it is priced at $3 per million input tokens and $15 per million output tokens, with a 1 million-token context window, rolling out today through the Gemini API and other surfaces.
Interpretation
Google released Gemini 4 Neon, a frontier model aimed at complex software engineering, knowledge work and science workflows, emphasizing fast end-to-end task completion and reporting both capability rankings and per-token pricing. Within the Gemini 4 family, Carbon handles the hardest long-horizon projects and Hydrogen handles fast everyday execution; Neon is positioned to balance reasoning depth and efficiency between them. At its xhigh thinking level, Neon scores 61 on the Artificial Analysis Intelligence Index, third overall behind larger, slower models, and ranks second on the Vals Index on the Pareto frontier.
In agentic settings Neon is built to run standalone or act as a high-judgment orchestrator directing parallel Hydrogen subagents in Google Antigravity. In long sessions it cuts redundant file reads, shares progress as it works, and handles context compaction reliably. Neon ranks first overall on the Artificial Analysis Coding Agent Index (58.6), scores 65.2% on Terminal-Bench 4.1 and 46.2% on FrontierCode 2.0 at its default high setting.
Neon's four thinking levels scale steadily, letting teams dial intelligence and token spend to each workload. At its default high level it pairs a high Artificial Analysis Intelligence Index score with a 32% lower cost per task than Carbon. It sits on the Pareto frontier on both the Vals Index and Agent Arena at high, with roughly 50% to 80% lower cost per task than top competing frontier models; xhigh gives its highest intelligence score while still costing 36%+ less per task than competing models at their highest setting.
Perspective
The demonstrated scope is the model's capability and pricing at launch: 61 on the Artificial Analysis Intelligence Index at xhigh, second on the Vals Index, 1336 on the Agent Arena Index, and scores on coding, science and cybersecurity benchmarks; the intended readers are developers and enterprise teams evaluating model choices. Rollout covers the Gemini API paid tier, Google AI Studio, Android Studio, Gemini for business, Google Antigravity and the Gemini app, with cybersecurity defenders in the Fairwind Program receiving access over the following days.
Several conclusions come from internal Google testing, such as the lowest fabrication rate yet in internal Google Antigravity testing, the 2.45% shortcut rate on the coding cheating evaluation, and the 43%–56% reduction in under-service and 15%–35% reduction in overreach on the agentic behavior evaluation; these figures correspond to internal evaluation settings. The 0.0% attack success rate on Gray Swan's indirect prompt injection benchmark was reached at the highest thinking level with real-time safeguards.
