Google introduced Gemini 4 Neon on November 18, 2026, placing it between the fast everyday-focused Hydrogen and the heavier Carbon model. For users choosing a model, the key questions are who can access Neon, which routes are free, and how subscription access differs from the token-metered API. Google says Neon is rolling out to Google AI Pro and Ultra subscribers in the Gemini app and Antigravity, and to developers on the paid Gemini API. At our November 21 source check, its standard API rate was $3 per million input tokens and $15 per million output tokens. The November 21 API rate card lists no free tier for Neon. Google launch · Google API pricing
Where can you try Neon?
| Route | Google's announced availability | What to check |
|---|---|---|
| Gemini app | Google AI Pro or Ultra subscribers can choose Neon from the model selector, during rollout | Check your account, country and current selector; an announcement is not a guarantee that every account has switched |
| Google Antigravity | Eligible paid Pro/Ultra users; Google also lists business access | Look for the Neon model entry. Development-agent quota or billing is not shown to be identical to the API billing schedule |
| Gemini API / Google AI Studio | Paid developer tier; model ID gemini-4-neon |
Confirm an enabled billing project and published rate limits; there is no Neon free API tier |
| Gemini for business | API and model dropdown paths for supported enterprise users | Enterprise terms and actual account provisioning may differ |
AI Studio has another access route: the launch also says Pro and Ultra subscribers can try Neon in AI Studio. That subscription entitlement does not establish free token-metered API calls.
Google's announcement mentions a cybersecurity-specific route for eligible Fairwind participants over subsequent days; that is not a general consumer access promise. Neither the launch announcement nor our research establishes that Neon appears in every region at the same minute. Launch · Model card
What do you pay on the API?
These are USD per one million tokens on Google's paid Gemini API, not the monthly price of Google AI Pro or Ultra. The following Standard rate-card snapshot was checked on November 21, 2026:
| Model, standard paid API | Input / 1M | Output / 1M | Important condition |
|---|---|---|---|
| Gemini 4 Hydrogen | $1.50 | $7.50 | Fast everyday-work option; different consumer/free API access rules |
| Gemini 4 Neon | $3.00 | $15.00 | Paid tier only; output pricing includes thinking tokens |
| Gemini 4 Carbon | $2.00 | $10.00 | Temporary introductory price through Nov 30, 2026 |
| Gemini 4 Carbon | $4.00 | $20.00 | Scheduled rate from Dec 1, 2026 |
At the November 21 check, this produces a counterintuitive comparison: Neon's $3/$15 nominal rate is 50% higher per token than Carbon's temporary $2/$10 rate. If the listed schedule takes effect on December 1, Neon becomes 25% lower per token than Carbon's $4/$20 standard rate. This is a rate-card comparison, not proof of the cheapest completed job. A model that produces more reasoning and output tokens could still cost more for your particular task. Google advertises 32% lower per-task costs than Carbon in its selected default-high benchmarks, which should be read as a vendor-reported evaluation rather than a guarantee for your prompts. Always recheck the live rate card before a purchase or deployment.
For Neon, Google lists standard cached input at $0.15 per million, described as implicit caching only on the official pricing page. Batch and Flex have lower listed per-token rates ($1.50 input / $7.50 output per million), while Priority is listed at $5.40 / $27.00. These are different processing modes, not evidence that all are automatically available on your account. Searches, maps, other tools or enterprise arrangements can have separate terms. API pricing
A simple illustration, not a billing quote: 100,000 uncached input tokens and 20,000 billed output tokens on Neon's standard rate would be $0.30 + $0.30 = $0.60, excluding extras and any difference between your intended output and actual billable thinking tokens. The billing metric is tokens actually processed, so estimate the output and check usage records instead of assuming a fixed cost per prompt.
What does Neon do differently from Hydrogen and Carbon?
Google positions Hydrogen for fast everyday tasks, Carbon for the most difficult long-horizon work, and Neon for coding, analysis, and tool-heavy tasks that need a balance between capability and efficiency. This is Google's positioning, not a universally tested winner for every use case. The Neon model card states that Neon is adapted from Carbon, supports text, image, audio and video inputs, and uses a combined context window up to one million tokens. It specifies a default synchronous output limit of 64,000 tokens, with separately described background long-decode capability; a one-million-token context does not mean an ordinary chat reply can always be one million tokens long. Google model card
Google highlights long coding sessions, debugging, multi-step research and enterprise document workflows. For example, it describes Neon as an orchestrator of parallel Hydrogen subagents in Antigravity. This is a named workflow supported by that product, not proof that a regular Gemini chat always spawns subagents or uses that architecture. Your own choice should hinge on whether the model's actual results, latency and total token cost improve the job you need done. Google launch
Do the leaderboards prove it is faster and better?
No single benchmark proves that. The figures below are an Artificial Analysis snapshot checked November 21, 2026. Google's November 18 launch cites an Artificial Analysis Intelligence Index score of 61 at the xhigh reasoning setting, as well as its own coding and agent evaluations. Independent Artificial Analysis's model page also reports a score of 61 at xhigh, but notes that the model is slow and verbose in that benchmark setting: it reports 21.5 minutes per Intelligence Index task, against a 12.7-minute comparator median at our November 21 check. That is a specific benchmark task duration, not the expected time for an ordinary chat reply.
A number labeled “#3” in a launch post and a later rank displayed on a live leaderboard can differ because rankings, model pools, settings and dates change. If you cite either ranking, attach the platform and measurement date instead of presenting it as a permanent fact. We did not perform an independent task-for-task test for HotRank, and Google partner testimonials are promotional statements, not an in-house review. Google launch · Artificial Analysis
Should you switch?
If you mainly ask quick questions or write short drafts, start by seeing whether Hydrogen already handles them well; higher nominal pricing does not automatically justify Neon. If you code across many files, analyze long mixed-format material or orchestrate multiple tools, Neon is worth a controlled comparison against your current option. If you primarily use the API, check the temporary Carbon pricing date first. If you have a free API project, Neon is not the drop-in free replacement.
A fair small test is to use the same task and inputs across models, record output quality, errors, time to usable completion, input/output tokens and total billed cost, then repeat on several representative tasks. Don't use Google's marketing demonstrations as your own test results. Also decide whether the task involves sensitive business material: check your organization's data processing terms, permissions and retention settings before uploading it.
How to get started without confusing products
- Gemini app: sign in to the Google account that holds your eligible Pro/Ultra plan, open the model selector and look for Gemini 4 Neon. If absent, note your region and rollout status; we cannot verify your account remotely.
- Developer API: use Google AI Studio, inspect the published paid-tier billing configuration and choose model
gemini-4-neonin the supported endpoint. Check limits and pricing separately from your consumer subscription. - Antigravity: in your eligible paid account, check the model menu and orchestration features against product documentation. Do not assume model use is unlimited because the API has a published rate.
- Before a large job, send a small representative test and inspect token use and the actual charge. If the model is missing, do not treat the news post as account-level availability evidence.
What remains uncertain
We have not confirmed regional availability for every account, actual per-account weekly caps, real-world comparative accuracy/latency on your workloads, or any default consumer quota equivalence between the Gemini app, Antigravity and the token-metered API. The model card also acknowledges hallucinations. Prices and access can change. Prices, access statements and benchmark figures are a November 21, 2026 source snapshot. Carbon’s December price is a published schedule, not confirmation that a future change has occurred.
Sources: Google's November 18 launch · Google API pricing · Google model card · Artificial Analysis.
For the earlier everyday model and Orchestrator announcement, read our Hydrogen access and pricing explainer. See more AI reporting.
Visual: the Gemini brand wordmark, not a Neon interface screenshot. Gemini and Google are trademarks of Google LLC. HotRank is not affiliated with or endorsed by Google.
