Inter-office memo
- To
- Everyone using Antigravity, the Gemini app or the Gemini API
- From
- Mo, Chief Executive
- Date
- November 18, 2026, 3:11 PM CST
- Re
- Gemini 4 Neon
Gemini 4 Neon is now Antigravity’s default on Pro and Ultra. Who gets it, and the cost
Google released Gemini 4 Neon on November 18, 2026, between Gemini 4 Hydrogen and Carbon, and made it the default for new Antigravity conversations on paid Google AI Pro, Ultra and Enterprise. Pro and Ultra subscribers can choose it in the Gemini app, and the API charges $3 per million input tokens and $15 output, paid tier only. Free and Google AI Plus accounts don’t get it.
What Google released on November 18
On November 18, 2026, Google released Gemini 4 Neon, the fourth Gemini 4 model after Argon, Carbon and Hydrogen. It arrived first as one line in the Gemini API release notes, “Gemini 4 Neon (gemini-4-neon) generally available (GA),” with no announcement. By that evening, Google had announced it on its blog, made it the default model in Antigravity on paid plans, and put it in the Gemini app for Pro and Ultra subscribers, and GitHub had added it to Copilot.
Google describes Neon as “bringing frontier-level intelligence with fast end-to-end task completion across complex software engineering, knowledge work, and science workflows,” and places it between the other two: “While Carbon tackles your hardest, long-horizon projects and Hydrogen handles fast everyday execution, Neon balances depth and efficiency.” On the API it has one 1,048,576-token window shared by input, thinking and output, a 65,536-token default for each response, and four thinking levels, low, medium, high and xhigh, starting at high. Google Cloud’s model page gives its knowledge cutoff as March 2026.
The name leaked first. TestingCatalog reported that “A reference to the model, listed as gemini-4-neon, appeared on the Gemini web app on November 16 in the upsell section,” and that on November 18 it “showed up in the Google Cloud SKU catalog” at the price Google’s docs now give.
Where you can use it
Here is where Neon was on November 19, by each company’s own pages:
| Where | Neon | Who gets it |
|---|---|---|
| Antigravity app, CLI and IDE extensions | Yes, the default for new conversations | Paid Google AI Pro, Ultra and Enterprise, and Pro trials “with eligible premium devices” |
| Gemini app | Yes, in the model drop-down | Google AI Pro and Ultra |
| Gemini API | Paid tier only | “not available on the Free tier” |
| Google AI Studio | Yes | Paid API projects, and Google AI Pro and Ultra subscribers |
| Gemini Enterprise Agent Platform, formerly Vertex AI | GA | Google Cloud customers, in the global, us and eu locations |
| GitHub Copilot | GA, rolling out gradually | Copilot Pro+, Max, Business and Enterprise |
| Cursor | Listed as a third-party model | Premium and above, from prepaid credit |
| Free and Google AI Plus plans | No | Not in Antigravity or the Gemini app |
In the Gemini app, Google’s plans page now lists “Access to Gemini 4 Neon” on Google AI Pro, at $19.99 a month, and “Highest access to Gemini 4 Neon” on Ultra, and nothing for Plus. Google says Tasks in Gemini “can now run on Neon.” The app’s limits page gives no per-model numbers: usage is “compute-based,” and Pro gets “4x higher than standard limits.” For developers, Google adds Android Studio on the paid API tier, and says “all Pro and Ultra subscribers can try it in AI Studio.”
Neon in Antigravity
Antigravity is where most people will notice Neon, because it changed the default. Google’s models page says Neon “appears at the top of the model selector on every plan that includes it. It’s the default model for new conversations if you haven’t chosen a model; if you’ve already selected one, Antigravity keeps your selection.” The launch post says it’s rolling out “across paid Google AI Pro and Google AI Ultra subscribers in the Antigravity desktop app, Antigravity CLI, and Antigravity IDE extensions,” and that “Enterprises can access Gemini 4 Neon in Google Antigravity through Gemini for business.”
Google’s forum announcement sets out who gets what:
| Plan | Gemini 4 Neon | Orchestrator (Neon) |
|---|---|---|
| Free and Google AI Plus | Not included | Not included |
| Google AI Pro, paid | Included | Included |
| Google AI Pro, trial | Only on trials with eligible premium devices | Only on trials with eligible premium devices |
| Google AI Ultra | Included | Included |
| Enterprise | Included through Gemini for business | Not included |
Three things differ from the API. Neon’s context window in Antigravity is “Up to 512K tokens,” half the API’s. Its levels are named Low, Medium, High, the default, and Extra, which is xhigh in the CLI: --model gemini-4-neon-xhigh or --effort xhigh. And it comes with Orchestrator (Neon), which uses Neon “for upfront planning, architecture, and complex reasoning while handing off routine code edits, package updates, and test runs to Hydrogen,” and goes back to Neon “If Hydrogen hits friction.” Orchestrator “runs with a fixed configuration, so you can’t choose a thinking level for it,” and the models page still calls it the “recommended default” for “Most day-to-day development.” Ultra now has three Orchestrators, Neon, Carbon and Argon; paid Pro has Neon and Argon.
On quota, Google’s announcement says Neon at High has “about 32% lower cost per task than Carbon, so your quota goes further.” The plans page gives no figures per model; it says limits “correlate with the amount of work done by the agent.”
Not everyone has it working. Under that announcement, a Google AI Ultra subscriber wrote that on Neon’s Extra level “the first message fails with: INVALID_ARGUMENT (code 400): Thinking level is unsupported: THINKING_LEVEL_XHIGH.” Other Ultra subscribers have reported the same error on Hydrogen’s Extra level, which they’ve called unavailable since November 16. The same post points to a thread where Ultra users say Gemini 4 models have answered with little or no thinking since November 15. Google hadn’t replied in either by early November 19. Our memo on Carbon’s speed covers those reports.
What it costs
Neon “uses flat pricing across the full 1M context window.” Here it is beside the other Gemini 4 models on the API’s pricing page, in US dollars per million tokens:
| Model | Free tier | Input | Output, thinking included | Cached input | Thinking levels |
|---|---|---|---|---|---|
| Gemini 4 Hydrogen | Yes | $1.50 | $7.50 | $0.075 | Up to xhigh, high by default |
| Gemini 4 Neon | No | $3 | $15 | $0.15 | Up to xhigh, high by default |
| Gemini 4 Argon or Carbon, through November 30 | No | $2 | $10 | $0.10 | Up to high, medium by default |
| Gemini 4 Argon or Carbon, from December 1 | No | $4 | $20 | $0.20 | Up to high, medium by default |
Batch and Flex requests cost half. Caching is implicit only: Google discounts repeated input “by 95%” on prompts of at least 4,096 tokens, and Neon doesn’t support explicit caches. On the paid tier, what you send isn’t used to improve Google’s products.
Per token, Neon costs twice what Hydrogen does, and more than Carbon until Carbon’s introductory price ends on November 30. Per task, it can cost less sooner. Google says Neon at high has “a 32% lower cost per task than Carbon,” and Artificial Analysis’s leaderboard on November 19 put Neon at $6.44 a task at xhigh, against $7.76 for Carbon at high.
Other apps charge the same rates. GitHub’s pricing table bills Neon at $3, $0.15 and $15 per million tokens, in AI credits worth a cent each. Cursor says Neon “isn’t part of the usage included with Cursor plans” and bills it from prepaid credit.
If you pay for Google AI Pro or Ultra, your plan’s monthly Google Cloud credits can go toward “Gemini API usage for prompting, building, and deploying AI-powered apps from Google AI Studio,” Google said on November 13, after “a $5 minimum prepayment and balance.” Our memo on those credits covers how to claim them.
How good it is
Google published its numbers with the announcement. It says Neon ranks “#3 overall with a score of 61 on the Artificial Analysis Intelligence Index at its xhigh thinking level,” and “#1 overall on the Artificial Analysis Coding Agent Index (58.6),” with 65.2% on Terminal-Bench 4.1 and 46.2% on FrontierCode 2.0 at high. On SWE-sweep it reports 12.8% in the mini-SWE harness and 39.0% in Antigravity. Outside coding, it claims #2 on the Vals Index at 70.9% and #1 on Zapier’s AutomationBench at 56.3%.
Artificial Analysis’s own leaderboard bears out the headline. On November 19 it listed:
| Model | Intelligence Index | Cost per task | Minutes per task |
|---|---|---|---|
| Claude Fable 5.2, max | 65 | $26.54 | 41.8 |
| GPT-6.2 Astra, max | 62 | $4.00 | 13.9 |
| Gemini 4 Neon, xhigh | 61 | $6.44 | 21.5 |
| Gemini 4 Carbon, high | 60 | $7.76 | 37.1 |
| Claude Opus 5.5, max | 59 | $10.04 | 25.2 |
| Gemini 4 Hydrogen, xhigh | 54 | $3.16 | 15.3 |
So Neon edges Carbon for less money and time, but GPT-6.2 Astra scores a point higher, faster and cheaper still. Google’s model guide keeps Carbon as “Our strongest reasoning model” for “system architecture, multi-file debugging, dense multimodal analysis, formal proofs, and security audits,” and Neon for work “where Hydrogen isn’t enough and you don’t need Carbon’s extra verification.” On safety, Google says Neon “took shortcuts on just 2.45% of tasks on our coding cheating evaluation (the lowest rate of any Gemini 4 model).”
What to do now
- In Antigravity, check which model a new conversation starts on. On paid Pro, Ultra and Enterprise it’s Neon unless you chose another. Try Orchestrator (Neon) for everyday work, Google’s recommended default, and keep Carbon for the hardest jobs on Ultra.
- If Extra fails, use High. Until Google answers the reports of “Thinking level is unsupported,” High is Neon’s default and the level Google’s 32% cost claim is based on.
- On the API, try it where Hydrogen falls short. You need a paid-tier project. Google’s advice is to start at
high, keepxhigh“for hard, multi-step problems wherehighfalls short,” and run your tests at every level “with a highmax_output_tokens.” - Check the migration list before you switch. From another Gemini 4 model, change the ID, but note Neon defaults to
highwhere Carbon defaults tomedium. From Gemini 3 models, removetemperature,top_p,top_kand explicit caching, raise lowmax_output_tokenscaps to 65,536, and set HTTP timeouts “to at least 15 minutes.” Neon “rejectsminimalwith an error.” - On
gemini-3.1-pro-preview, plan the move. Google’s model guide namesgemini-4-neonas its replacement, “orgemini-4-carbonfor your hardest problems,” and the deprecations page still gives 3.1 Pro Preview no shutdown date. - Watch what agents spend. Google bills “every token the model processes, including thinking tokens,” at $15 per million on output, and a coding agent runs many turns for one task.
What Google hasn’t said
- How much Antigravity quota or Gemini app usage a Neon turn takes, beside Carbon and Hydrogen.
- Why Antigravity caps Neon at 512K tokens when the API offers 1M.
- Whether Neon comes to Free or Google AI Plus, in Antigravity or the Gemini app.
- What’s behind the Extra-level errors and missing thinking that Ultra subscribers report on its forum.
- Whether other apps get Orchestrator. Google documents it in its own app and CLI.
Neon in Enjoy
Enjoy runs Antigravity on your Google account, through the server Google offers apps such as Zed and JetBrains IDEs, beside Claude Code, Codex and the other labs’ agents, each on its own plan. Its model list shows “the Gemini models Google offers your account,” fetched from the agent when you open its settings, so when Google offers your account Neon through that server, it appears in the model list without an Enjoy update. On Free and Google AI Plus it won’t, because Google doesn’t offer Neon there. When Neon stalls, you can switch agents in the middle of the conversation, and Claude Code or Codex picks the job up on the plan you already have.
OpenCode in Enjoy uses the providers you set up in it, so it can reach the Gemini API with your own key. By November 19, models.dev, the catalog OpenCode reads its model list from, listed Neon with its 1,048,576-token context and $3 and $15 prices, and OpenRouter offered it too. You pay Google’s or OpenRouter’s price either way. Enjoy is a monthly subscription, with the first seven days free (pricing). It doesn’t add Neon to an account or a plan Google hasn’t given it to.
We’ll update this memo when Google publishes Neon’s quotas, answers the Extra-level reports, or brings Neon to more plans.
9.Questions.
Is Gemini 4 Neon in Antigravity?
- Yes, since November 18, 2026, on paid Google AI Pro, Ultra and Enterprise, and on Pro trials “with eligible premium devices.” It’s the default model for new conversations, with up to 512K tokens of context in Antigravity, half the API’s 1M. Free and Google AI Plus accounts don’t get it, and Orchestrator (Neon) is on Pro and Ultra but not Enterprise.
Why did Antigravity switch me to Gemini 4 Neon?
- Because Neon is now the default when you haven’t chosen a model. Google’s models page says “if you’ve already selected one, Antigravity keeps your selection,” so only new conversations without a choice start on Neon. Choose another model in the selector under the prompt box, or with `--model` in the CLI.
Is Gemini 4 Neon free?
- No. The Gemini API’s pricing page says “Gemini 4 Neon is not available on the Free tier,” and Antigravity and the Gemini app offer it on Google AI Pro and Ultra, not on Free or Google AI Plus. On the API’s paid tier it costs $3 per million input tokens and $15 per million output, thinking included.
Is Gemini 4 Neon better than Gemini 4 Carbon?
- By Artificial Analysis’s Intelligence Index, slightly, and it’s faster. On November 19, its leaderboard put Neon at 61 at `xhigh` and Carbon at 60 at `high`, with Neon averaging 21.5 minutes and $6.44 a task to Carbon’s 37.1 minutes and $7.76. Google still calls Carbon “Our strongest reasoning model” and keeps it for the hardest long-horizon work.
Can I use Gemini 4 Neon in GitHub Copilot or Cursor?
- Yes, in both, at Google’s API prices. GitHub made it generally available on November 18 for Copilot Pro+, Max, Business and Enterprise, billed at $3 and $15 per million tokens in AI credits, rolling out gradually. Cursor’s docs list it as a third-party model paid from prepaid credit on Premium and above, and say “Plus and Free don’t include third-party models.”
Does Gemini 4 Neon use less Antigravity quota than Carbon?
- Google says so, without numbers. Its launch post says Neon at its default High level shows “a 32% lower cost per task than Carbon,” and its forum announcement adds “so your quota goes further.” Antigravity’s plans page doesn’t say how much quota a Neon turn uses.