Google has launched Gemini 4 Neon, an artificial intelligence model engineered for demanding knowledge tasks such as software engineering and multi-step scientific workflows. Koray Kavukcuoglu, vice president at Google DeepMind and Google senior AI architect, detailed the technical specifications and rollout plans on the official Google blog. The new model is specifically targeted at reducing latency and operational expenditures in developer environments where agent workflows execute frequent tool calls and multi-turn reasoning loops. API pricing is set at $3 per million input tokens and $15 per million output tokens across a one-million-token context window. To support heavy iterative pipelines, Google introduced prompt caching that slashes input costs by 95 percent to $0.15 per million tokens, allowing development teams to keep entire codebases active in context at minimal expense.

In terms of unit economics and computational efficiency, Gemini 4 Neon reduces task processing costs by 32 percent relative to Gemini 4 Carbon, driving per-task expenses down to 68 percent of previous baseline levels while matching or exceeding flagship coding capabilities. Google has released several models since October, including Argon, Carbon, and Hydrogen. While Carbon demonstrated high reasoning ability, its pricing and latency made it challenging to sustain in production agents requiring dozens of iterative cycles. Benchmark evaluations show that Neon scored 61 points on the Artificial Analysis Intelligence Index at its highest reasoning tier, placing third overall, and earned 58.6 points on the Artificial Analysis Coding Agent Index to take first place. Under Google Antigravity testing, its mini SWE harness sweep score advanced from 12.8 percent to 39.0 percent, alongside a 99.2 percent pass rate on agent honesty tests during external tool execution.

Google has initiated immediate deployment across its product ecosystem. Gemini 4 Neon replaces Argon in the model selector for paying Gemini subscribers. In parallel, Google has begun distributing Gemini 4 Neon Cyber, a specialized defensive variant designed for security operations, to verified defenders through its Fairwind program. For engineering organizations seeking to scale autonomous multi-step agents in production workflows, the combination of a one-million-token context window, top-tier coding performance, and a 32 percent reduction in per-task computing costs establishes a new economic baseline for enterprise model adoption.