Gemini 4 Neon: Google’s Carbon-Level AI at Flash-Like Speed

Google has unveiled Gemini 4 Neon, a new model positioned between Carbon and Hydrogen in its Gemini 4 lineup, delivering performance matching Carbon on demanding tasks but executing at Flash-level speeds. The immediate change affects the subscription model tiers: Neon replaces Argon, and 3.1 Pro is being retired. Meanwhile, 3.5 Flash-Lite will become “Auto” in the model picker for all paid users.

Previously, the Gemini 4 family consisted of three models: Hydrogen for everyday queries, Argon for coding and general knowledge work, and Carbon for the most complex, intelligence-testing applications. Neon now sits in the deep analysis tier alongside Carbon, offering Carbon-level capabilities on coding, reasoning, and agentic tasks, but with latency and costs that better match the Flash experience.

Benchmark Performance and Comparisons

Neon has distinguished itself across a broad swath of benchmarks. It scores 61 on the Artificial Analysis Intelligence Index at the “xhigh thinking” level, making it third overall, just behind certain larger but slower AI models. It tops the steerability ranking on the Agent Arena Index, proving adept at adapting to mid-task corrections. On both the Legal Research Bench and Tax Agent Bench, Neon emerges first, showing particular strength in specialized professional workflows.

Across additional evaluation tools like Surge AI and professional benchmarks involving finance, accounting, spreadsheet modeling, and complex constraint-following, Neon is either the top performer or among the very best. Its performance on visual metrics is strong too: it scores 73.5% on Chartography when interpreting dense professional charts. It also fares well on long-horizon business simulations: it holds the No. 1 ranks on Zapier’s AutomationBench and Vending-Bench 2, and rates highly in global agent session evaluations like the Agent Arena Index.

Pricing and Subscription Model Shifts

Neon is now available to AI Pro and AI Ultra subscribers. Under promotional pricing through November 30, Argon and Carbon are priced at $2 per million input tokens and $10 per million output tokens. After that date, those two will double to $4 input / $20 output per million tokens. Neon is priced 25% lower than Carbon post-promotion, while being twice as expensive as Hydrogen now.

The model picker has been reshuffled: “Auto” adapts to your use-case; Hydrogen remains the go-to for everyday assistance; Neon handles deeper analytic tasks; and Carbon is positioned for code and complex workflows.

“Gemini 4 Neon launches: Carbon quality at Flash speeds, for AI Pro” marks a strategic move. By slotting Neon between Carbon and Hydrogen, Google aims to give users the power of its strongest model without the full cost or latency overhead.

Analytical Perspective: The introduction of Neon signals Google’s recognition that many users need high-caliber AI reasoning but can’t tolerate the latency or cost of top-tier models like Carbon. By delivering Carbon-level intelligence at Flash speeds and a lower price, Google is reducing the trade-off between performance and resource usage. It also tightens its competitive posture in the AI market, where efficiency and cost-effectiveness are becoming just as important as raw capability. Watch closely for how users adopt Neon in real workflows—its success may pressure rivals to optimize their mid-high tier models similarly.