Google released Gemini 4 Neon on November 18, 2026, pricing it at $3 per million input tokens and $15 per million output tokens. The mid-tier model is live in the Gemini app for paying US subscribers, and developers can call it through the Gemini API and Google AI Studio the same day. Three days later, the pricing and benchmark claims are fresh enough that teams are deciding now whether to switch.
What Google launched on November 18
Neon sits between Gemini 4 Hydrogen and Gemini 4 Carbon in Google's lineup. According to 9to5Google, Google says Neon matches or beats Carbon on coding, reasoning and agentic tasks while running at Flash-level speed and costing less. Those are Google's own claims, and independent testing is still early.
The timing matters. OpenAI released GPT-6.2 Astra and GPT-6.2 Sol on November 17, according to AI Release Tracker. Google's benchmark chart compares Neon against Claude Opus 5.5 and GPT-6.2 Sol, so the launch is aimed squarely at the models developers are already weighing.
Google has now released four Gemini 4 models. Silicon AI News reports that Carbon's launch discount ends November 30, a deadline that gives budget owners a reason to rerun their numbers this month.
Where Gemini 4 Neon is available now
Neon rolled out the same day across several Google products, according to Google's announcement:
- Gemini app: AI Pro and Ultra subscribers can pick Neon from the model menu.
- Gemini API and Google AI Studio: developers can call the model directly.
- Gemini for business: enterprise customers can reach Neon through the API or the model drop-down.
- Google Antigravity: the model is available here as well.
Neon supports a 1 million token context window and four thinking levels: Low, Medium, High and Extra High. The thinking levels let you trade speed and cost for depth on each request, which matters when the same prompt runs at scale.
The app change is the one most people will notice first. 9to5Google reports that Neon replaces Gemini 4 Argon in the Gemini app's model menu, but Google's own post does not say so. Check your menu to see which models you actually have.
Google also says a dedicated Neon for cyber version will reach vetted security defenders through its Fairwind Program “in the coming days.” Google has not said which countries get access to Neon, so the rollout outside the United States remains unclear.
Pricing: $3 at launch, with conflicting post-promotion figures
Launch pricing is the number most teams will budget around. Neon costs $3 per million input tokens and $15 per million output tokens, according to CASRAI and 9to5Google.
The next figure is harder to pin down. 9to5Google says Neon rises to $4 input and $20 output after promotional pricing, and that it is 25 percent cheaper than Carbon on that basis. GSMArena lists $4 and $20 as Carbon's price. Those two figures do not fit together cleanly. The 25 percent gap matches $3 against $4, so it appears to compare Neon's launch price with Carbon's $4 rate. If Neon settles at $4 and $20, its price advantage over Carbon disappears.
| Model | Input per million tokens | Output per million tokens | Source |
|---|---|---|---|
| Gemini 4 Neon, launch | $3 | $15 | 9to5Google, CASRAI |
| Gemini 4 Neon, after promotion (reported) | $4 | $20 | 9to5Google |
| Gemini 4 Carbon | $4 | $20 | GSMArena |
| GPT-6.2 Sol | $2 | $10 | AI Release Tracker, unverified |
| GPT-6.2 Astra | $8 | $40 | AI Release Tracker, unverified |
On list price, GPT-6.2 Sol is the cheapest option in this comparison. Neon's case therefore rests on quality and speed rather than on being the lowest-cost model. The OpenAI figures come from a secondary tracker and have not been checked against OpenAI's own pricing.
Before you budget, check Google's pricing page for current Neon and Carbon rates. The post-promotion figures are the ones most likely to change.
How Neon's benchmarks compare with Carbon and GPT-6.2 Sol
Google's headline claim is that Neon matches or beats Carbon on several tests. The third-party numbers are thinner, but a few are in circulation.
On the Artificial Analysis Intelligence Index, Neon scores 61 at the Extra High setting and ranks third overall, according to 9to5Google. The same report puts Neon second on steerability in the Agent Arena Index. The benchmark tracker Priors measured output at about 110 tokens per second at the xhigh setting, a third-party reading reported by VentureBeat.
The Digital Brief reports a Coding Agent Index score of 58.6 and first place at launch. That figure has not been confirmed against Google's own post.
Two quotes add a human view. Thai Tran, Glean's AI product lead, said Neon “matched or exceeded Gemini 4 Carbon on overall quality” in early tests, a line VentureBeat attributes to Google's announcement. On launch day, Varun Mohan wrote on X: “It's cheaper, faster, and better behaved than Carbon.” The source does not give Mohan's role, and a social post is a reaction rather than a measurement.
Google describes Neon as offering “frontier-level reasoning for complex, high-frequency workflows,” a phrase CASRAI quotes from the official launch description. That positioning signals who Neon is for: repetitive, high-volume work where cost per call matters as much as top scores. According to The Neuron, Neon's parameter count and training compute were not disclosed.
Launch charts deserve the same scrutiny whichever lab publishes them. METR's report on Anthropic came with most of its evidence redacted, a reminder that outside verification matters. For Neon, the strongest evidence so far is that several independent trackers place it near the top, not any single chart from Google.
What to do if you use Gemini or build on the API
- Open the Gemini app's model menu if you have AI Pro or Ultra and select Neon. Run five or six of your real prompts through Neon and your current model, then compare the outputs side by side.
- Test the 1 million token context window with the longest document you handle. Try each thinking level on one hard task to see whether Extra High justifies its cost.
- If you build on the API, estimate monthly spend at $3 and $15 per million tokens, then rerun the estimate at $4 and $20. Verify both rates on Google's pricing page before you commit.
- If you run Carbon, note November 30, the date its launch discount ends, and plan your migration or budget around it.
- Enterprise teams using Gemini for business can add Neon through the API or the model drop-down. Confirm default-model changes with whoever manages your account before you switch.
- Security teams should check whether they qualify for Google's Fairwind Program when the cyber version opens to vetted defenders.
What to watch next
Several questions will decide whether Neon's pricing holds. The first is whether the post-promotion rate becomes official at $4 and $20, or whether Google keeps $3 and $15 longer. The second is when the Fairwind cyber version opens, since Google has only said “in the coming days.” Watch also which countries get access, and whether independent leaderboards keep Neon in the top three as more testers publish results.
Google is also pushing Gemini beyond chat. Its Gemini Tasks feature now calls businesses and controls Mac apps in the US, which makes per-token pricing matter more when an agent runs without someone watching. A model that is cheaper and better behaved on repetitive work is the kind of model that gets left running. Whether OpenAI responds on price in the coming weeks will show how much of this competition comes down to cost.




