Google released Gemini 4 Neon on November 18, 2026, its fourth Gemini 4 model in seven weeks, and the new middle tier arrives with a specific pitch: frontier-level reasoning at a lower cost per task than the flagship Carbon. Neon is priced at $3 per million input tokens and $15 per million output tokens, with a 1 million-token context window and four thinking levels. Independent evaluator Artificial Analysis puts it at 61 on its Intelligence Index at the xhigh setting, third overall and one point ahead of Gemini 4 Carbon at high. For teams that run a lot of agent work, the benchmark score matters less than the cost of finishing a job.
The Fastest Release Cycle in Google's History
Google's Gemini 4 family now spans four models in about seven weeks. Argon arrived on September 30, Carbon on November 10, Hydrogen on November 12, and Neon on November 18. Artificial Analysis, which tracks release cadence, called Neon Google's fourth model in seven weeks. The pace matters because each release targets a different kind of work. Hydrogen handles high-volume everyday tasks. Carbon targets long-horizon projects that test a model's limits. Neon sits between them, aimed at complex coding, analytical reasoning, and production enterprise workflows.
In the Gemini app, Neon replaces Argon under the "Deeper analysis" label, while Carbon stays at "Code & complex work" and Hydrogen covers everyday help. Google AI Pro and Ultra subscribers can select Neon directly. Developers get it through the Gemini API paid tier, Google AI Studio, Android Studio, and Gemini for business.
The Price and the Positioning
Neon costs $3 per million input tokens and $15 per million output tokens. Cached input runs at $0.15 per million tokens, a 95 percent discount. Carbon is currently discounted to $2 and $10 through November 30, after which it rises to $4 and $20. That promotion creates an odd window: until the end of November, Neon actually costs more per token than Carbon. Artificial Analysis notes that Neon only becomes cheaper per token on December 1. Google's argument is about total cost per completed task rather than the sticker price, and there the numbers favor Neon. At its default high thinking level, Google says Neon delivers a 32 percent lower cost per task than Carbon. Artificial Analysis measures it at $6.44 per task, about 17 percent below Carbon at standard pricing, and about 21 minutes per task against roughly 36 minutes for Carbon.
Benchmarks and the Cost-Per-Task Argument
Neon scores 61 on the Artificial Analysis Intelligence Index at xhigh, one point ahead of Carbon at high, and ranks behind GPT-6.2 Astra and Claude Fable 5.2. Google's blog lists a set of results: Neon ranks second on the Vals Index at 70.9 percent, first on the Legal Research Bench and the Tax Agent Bench, and first overall on the Artificial Analysis Coding Agent Index at 58.6. It scores 65.2 percent on Terminal-Bench 4.1 and 46.2 percent on FrontierCode 2.0 at its default high setting. On Arena's Agent Arena Index, which draws on more than 4,200 real-world agent sessions, Neon debuts at second globally with a score of 1,336.
Artificial Analysis found that Neon in the Antigravity CLI scores 59 on the Coding Agent Index, three points ahead of Carbon, and lands in line with Claude Code running Claude Fable 5.2 and Claude Opus 5.5 at a much lower cost per task. Google says Neon can also act as an orchestrator directing parallel Hydrogen subagents in Antigravity.
Built for Long-Running Agent Work
The design goal behind Neon is endurance. Google says the model cuts down on redundant file reads, shares progress as it works, and handles context compaction reliably in long sessions. Context compaction is the step where a long working history gets summarized so an agent can continue, and it is a step that has caused other models to drift. Neon's four thinking levels let a team dial reasoning depth to the workload, with xhigh reserved for the hardest engineering, analytical, and scientific problems.
Security and the Fairwind Program
Neon for cyber is rolling out to trusted defenders through Google's Fairwind Program. Google says Neon produces more secure code than Carbon across easy, medium, and hard challenges on AutoBaxBench, reaching state of the art on the hard tier at 76 percent. On WebExploitBench v1.2 it reaches 51.26 percent, and on SEC-bench Pro it scores 90.5 percent. On Collinear's CWE-bench v2 it reaches 70.7 percent, a 3.4 point uplift over Carbon. Google also says the model refuses only 1.98 percent of tasks on Terminal-Bench 4.0 and 0 percent on Terminal-Bench-Science, meaning it declines very little legitimate work in those areas.
What This Means for Your Stack
If you run agent workflows at volume, Neon's pitch is straightforward: get close to Carbon's reasoning quality at Flash-like speed and a lower cost per task. If you are already on Argon, Google has made Neon the upgrade target. The practical caution is the pricing window. Carbon is half price until November 30, so a token-by-token comparison during that period will mislead you. Compare cost per completed task instead, and run your own evals on the work you actually do.
The bigger story is the cadence. Four frontier models in seven weeks means the model you standardize on today will have a faster, cheaper sibling within a month. Build your prompts and evals so you can swap model IDs without rewriting your product.
Read more AI model breakdowns and monetization playbooks at crowe-endeavors.com.
Sources
"Gemini 4 Neon Is Here, And It's a Beast!" (Prompt Engineering (YouTube), November 18, 2026) — https://www.youtube.com/watch?v=Q0ShBenKyqo
"Introducing Gemini 4 Neon" (Google, November 18, 2026) — https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-4-neon/
"Google continues to accelerate its release cadence with Gemini 4 Neon" (Artificial Analysis, November 18, 2026) — https://artificialanalysis.ai/articles/gemini-4-neon-intelligence-gain
"Google launches Gemini 4 Neon with Carbon quality at Flash speeds" (9to5Google, November 18, 2026) — https://9to5google.com/2026/11/18/gemini-4-neon-announcement/