Gemini 4 Neon: $3/$15 Price, Benchmarks vs Carbon and Opus
Google launched Gemini 4 Neon on November 18, 2026 at $3 and $15 per million tokens. It is paid-tier only. Carbon is cheaper per token through November 30.

Google launched Gemini 4 Neon on November 18, 2026 at $3 and $15 per million tokens. It is paid-tier only. Carbon is cheaper per token through November 30.

On November 18, 2026, Koray Kavukcuoglu, SVP of Google DeepMind, published Introducing Gemini 4 Neon ↗. The model id on the Gemini API pricing page ↗ is gemini-4-neon. The Verge ↗ (Jay Peters, November 18) says the release follows Argon, Carbon, and Hydrogen. 9to5Google ↗ (Abner Li, November 18) covered the same launch. The picture above is an original graphic created by aitoolranked. We have not used Gemini 4 Neon.
Google's positioning sentence: "While Carbon tackles your hardest, long-horizon projects and Hydrogen handles fast everyday execution, Neon balances depth and efficiency."
The pricing page lists a 1 million-token context window and long-decode generation up to 1 million tokens. Standard rates are $3 per million input tokens, $0.15 cached, and $15 output. Output includes thinking tokens. Google's blog calls the cache price a 95% discount. Caching is implicit only. Neon is not on the free tier.
Google's launch post puts Neon on the paid Gemini API in Google AI Studio and Android Studio, in Gemini for business (the API and the model dropdown), in Antigravity for paid Google AI Pro and Ultra, and in the Gemini app dropdown for those subscribers. Tasks in Gemini can run on Neon. The Antigravity post ↗ (November 18) adds the desktop app, CLI, and IDE extensions, and enterprise access in Antigravity through Gemini for business. Eligible Fairwind Program members get Gemini 4 Neon for cyber over the next few days, Google says.
Orchestrator can pair Neon as planner with Hydrogen as executor in one session. Hydrogen runs routine edits, package updates, and tests, and hands back to Neon on friction. Thinking levels run from Low to Extra High. High is the default. 9to5Google names Low, Medium, High, and Extra High, and reports the app picker as Auto, Everyday help (Hydrogen), Deeper analysis (Neon), and Code and complex work (Carbon). That picker is 9to5Google's account.
Argon matches Carbon's input and output schedule. Rates below are per million tokens.
| Model | In | Out |
|---|---|---|
| Neon | $3.00 | $15.00 |
| Hydrogen | $1.50 | $7.50 |
| Carbon now | $2.00 | $10.00 |
| Carbon Dec 1 | $4.00 | $20.00 |
"Carbon now" runs through November 30, 2026. From December 1, Argon and Carbon move to $4.00 and $20.00, the step in our December 1 price note. Hydrogen stays at $1.50 and $7.50, with a free tier. Neon has none. Neon costs more per token than Carbon until November 30 and less from December 1. The Hydrogen note and the Carbon note cover those launches. The December 1 note records Claude Opus 5.5 at $4 and $20 on Anthropic's pricing page. Carbon matches that rate on December 1. Neon stays under it.
These figures are Google's claims from the launch post and the Antigravity post. We did not run them. Google says #3 on the Artificial Analysis Intelligence Index with 61 at xhigh, #2 on the Vals Index at 70.9%, and #1 on the Artificial Analysis Coding Agent Index at 58.6. It says 65.2% on Terminal-Bench 4.1 and 46.2% on FrontierCode 2.0 at default high. SWE-sweep is 39.0% in Google Antigravity against 12.8% on the mini-SWE harness. Arena's Agent Arena Index is #2 at 1336. Zapier AutomationBench is #1 at 56.3%. At default high, Google says cost per task is 32% lower than Carbon, and that internal Antigravity testing showed the lowest fabrication rate yet.
OrcaRouter ↗ (November 19), reading Artificial Analysis pages, says the tracker scores Neon 61.4 at xhigh and 59.6 at the default high setting, against Carbon at 60.0 at Carbon's maximum. It says Neon at low is 44.1 and Carbon at low is 54.2. OrcaRouter argues that the headline 61 is the least representative configuration, and says Google's docs tell developers to evaluate all four levels and pick the lowest that clears the bar. Artificial Analysis, on X, describes a ~17% lower cost per task than Carbon. That figure is the tracker's, not Google's 32%. We did not run the tasks.
"Google continues to accelerate its release cadence with Gemini 4 Neon, delivering a small Intelligence gain over Gemini 4 Carbon at a ~17% lower Cost per Task" — Artificial Analysis ↗ (@ArtificialAnlys)
"It outperforms Opus 5.5 and 6.2 Sol on most benchmarks, and beats Carbon on AA's Intelligence Index" — leo ↗ (@synthwavedd)
"We'll see if both models are benchmaxed or not." — Emirhan Erkan ↗ (@permaximum88)
"the flagship gets the launch post, the fast one does most of the actual work" — Heorhi Talochka ↗ (@heorhit)
"two model launches this week and the headline number was the price" and "both numbers come from the vendor." — norii ↗ (@noruii22)
"Gemini 4 Neon from @GoogleAI is live on AI Gateway today." — Vercel Developers ↗ (@vercel_dev)
Reddit was not reachable for this story.
Benchmark Neon this week if an agent loop already runs Claude Opus 5.5 or Gemini 4 Carbon. Until November 30, Carbon is cheaper per token. From December 1, Neon undercuts it. Price the finished task. Google's 32% and the tracker's ~17% are different claims. OrcaRouter's default-high score sits under Carbon.
Call high before xhigh. The headline 61 is the xhigh score. High is the default. Emirhan's question is the reason to run your own tasks.
Google AI Pro and Ultra get Neon in the app. A free API key does not. Hydrogen is the free-tier Gemini 4 model. Vercel ↗ says Neon is on AI Gateway. Tasks on Neon are in Tasks in Gemini. The agent surface is Google Antigravity.
The alternatives are already on this desk. Gemini 4 Carbon, Gemini 4 Hydrogen, Claude Opus 5.5 at the rate in the December 1 note, and GPT-6.2 Sol. Plan comparisons are in ChatGPT vs Claude vs Gemini. Coding seats are in the best AI for coding guide.
Verdict: For a paid agent loop, price the finished task, not the headline 61. A free API key cannot call Neon. Hydrogen is the free Gemini 4 option.