Key takeaways
- Gemini 4 Neon costs $3 per million input tokens and $15 per million output tokens; Argon costs $2 and $10 through November 30, then $4 and $20.
- Artificial Analysis scores Neon 60 and Argon 50 on its Intelligence Index at the high setting, with Neon finishing tasks in about half the time.
- Both have a 1M-token context window, no API free tier, and are in the Gemini app for Google AI Pro and Ultra subscribers.
Gemini 4 Neon vs Argon comes down to a date. Neon, which Google released on November 18, 2026, scores 10 points higher than Argon on the Artificial Analysis Intelligence Index at the same thinking setting and runs faster. It costs $3 per million input tokens and $15 per million output tokens, which is more than Argon today and less than Argon from December 1, when Argon's price doubles to $4 and $20.
For most new projects, Neon is the better pick. Argon is cheaper per token only until November 30, and on every test Google published for both models, Neon scores higher. This comparison is based on Google's announcements, model cards and pricing page, and on Artificial Analysis's published results. We did not test either model.
The two models in one paragraph each
Gemini 4 Argon came out on October 13, 2026, as the first public model of the Gemini 4 family. Google aimed it at deep, multi-step work in coding, finance, law, cybersecurity and science. Google's models page calls it "our efficient Gemini 4 frontier reasoning model", meaning it answers faster and uses fewer tokens per turn than Gemini 4 Carbon, the family's strongest model.
Gemini 4 Neon came a month later. Google's model card says Neon is adapted from Carbon and sits between Carbon, for the hardest long projects, and Hydrogen, for fast everyday tasks. Google pitches it at complex work that runs often: coding, agents, office tasks and science. Our news story on the Neon launch has the full announcement.
Gemini 4 Neon vs Argon at a glance
Prices are from the Gemini API pricing page and speeds from Artificial Analysis, both as of November 19, 2026.
| Gemini 4 Neon | Gemini 4 Argon | |
|---|---|---|
| Released | November 18, 2026 | October 13, 2026 |
| API model code | gemini-4-neon |
gemini-4-argon |
| Input price (per 1M tokens) | $3.00 | $2.00 through November 30, then $4.00 |
| Output price (per 1M tokens) | $15.00 | $10.00 through November 30, then $20.00 |
| Cached input (per 1M tokens) | $0.15 | $0.10, then $0.20 |
| Batch and Flex (input / output) | $1.50 / $7.50 | $1.00 / $5.00, then $2.00 / $10.00 |
| API free tier | No | No |
| Context window | 1M tokens | 1M tokens |
| Output limit | 64K by default, up to 1M | 64K / 1M |
| Knowledge cutoff | March 2026 | March 2026 |
| Artificial Analysis Intelligence Index | 60 (high), 61 (xhigh) | 50 (high), 48 (medium) |
| Output speed, high setting | 101 tokens/s | 65 tokens/s |
| Wait for first answer token, high | 40.64 s | 49.73 s |
| Gemini app | Google AI Pro and Ultra | Google AI Pro and Ultra |
How much Neon and Argon cost
As of November 19, 2026, Neon costs $3 per million input tokens and $15 per million output tokens on the Standard tier. The pricing page lists no change date for it. Batch and Flex, the two slower and cheaper ways to run requests, cost half: $1.50 and $7.50. Priority, for traffic that must not be delayed, costs $5.40 and $27.
Argon is still on an introductory price: $2 and $10 through November 30. From December 1 the page lists $4 and $20. Batch and Flex go from $1 and $5 to $2 and $10, and Priority from $3.60 and $18 to $7.20 and $36.
A concrete job shows the swap. Sending 1 million input tokens and getting 200,000 back costs $6.00 on Neon. On Argon it costs $4.00 today and $8.00 from December 1. Neon is 50% more expensive than Argon this month and 25% cheaper next month, on every tier.
The two models bill the same way otherwise. Cached input is 95% off for both, and both support only implicit caching, where Google detects repeated prompt prefixes on its own. Thinking tokens are billed as output. Neither has a free tier in the API. Grounding with Google Search gives 5,000 free requests a month, shared across Gemini 3 and newer models, then costs $14 per 1,000.
Argon also has a Contributor tier on the pricing page, an invite-only preview at $1 and $5 where Google may use your traffic to improve its products. Neon doesn't have one.
Context window and output length
On context, there is nothing to choose. Both read up to 1 million tokens, counting input and output together, and both take text, images, audio, video and PDFs. Google's models page lists the same "64K / 1M" output limit for both. Neon's model card explains the two numbers: up to 64K tokens in a normal request, and up to 1M through a background "long decode" mode.
What differs is how well each model uses a long prompt. On GraphWalks, Google's test of following links through 256K to 1M tokens of text, Neon scores 95.2% and Argon 84.2%. Both numbers are Google's own.
Speed: Neon finishes tasks in half the time
Artificial Analysis measured Neon at 101 output tokens per second at its high thinking setting, and 110 at xhigh, a higher setting Argon doesn't have. It measured Argon at 65 tokens per second on high and 58 on medium.
Both models think before they answer, so the wait for the first word of the answer is long. On high, Neon took 40.64 seconds and Argon 49.73 seconds. On Neon's medium setting the wait drops to 20.14 seconds, and it still scores 55 on the index, five points above Argon on high.
The clearest number is time per task. On the tasks in Artificial Analysis's index, Neon on high took 14.8 minutes on average and Argon on high took 29.7. For agents that run many steps in a row, that gap adds up.
Benchmarks: Neon leads every row Google published for both
The one independent score is the Artificial Analysis Intelligence Index v5.0.1, which combines many tests into one number. Read on November 19, it puts Neon at 60 on high and 61 on xhigh, and Argon at 50 on high. Google says Neon at xhigh ranks third overall on that index.
Google didn't put Argon in Neon's results. The Argon scores below come from the Gemini 4 Carbon model card, with results as of October 2026. Neon's come from its own model card, as of November 2026, except Vending-Bench 2, which is from the launch post. All of them are Google's own testing.
| Benchmark (Google's figures) | Gemini 4 Neon | Gemini 4 Argon |
|---|---|---|
| Vals Index v2.2 (finance, law, coding, tax) | 70.9% | 69.2% |
| Agents' Last Exam v1.1 (computer use) | 75.8% | 69.0% |
| AutomationBench (office automation) | 56.3% | 51.3% |
| Vending-Bench 2 (running a business, final balance) | $16,843 | $13,718 |
| RiemannBench (math) | 84.0% | 76.0% |
| Chartography (reading charts) | 73.5% | 71.6% |
| GraphWalks, 256K to 1M tokens | 95.2% | 84.2% |
| SEC-bench Pro (finding vulnerabilities) | 90.5% | 83.4% |
Neon leads on all eight. The gaps range from 1.7 points on the Vals Index to 11 points on GraphWalks. The coding tests don't line up: Google reported Argon on Terminal-Bench 4.0 and FrontierSWE 2, and Neon on Terminal-Bench 4.1 and FrontierCode 2.0, so those numbers can't be set side by side. Our Neon benchmarks article compares Neon with Carbon, Claude Opus 5.5 and GPT-6.2 Sol on the full table.
Argon has some results Neon hasn't been tested on, at least not publicly. Google reported 87% on ERQA, a test of reasoning about the physical world, 57% on RareBench for help with rare-disease diagnosis, and 52.2% on LabBench for lab reasoning. On Gray Swan's prompt injection test, where lower is better, Argon let 0.7% of attacks through and Neon, at its highest thinking setting, 0.0%.
Where you can use each one
Both models are on the Gemini API's paid tier, in Google AI Studio and Android Studio, and in Gemini for business. In the Gemini app, both are in the model menu for Google AI Pro and Ultra subscribers. Free users get neither.
There are three differences:
- Google Antigravity. Google's coding tool offers Neon to paid AI Pro and Ultra subscribers. Argon's announcement listed it for Ultra subscribers only.
- AI Mode in Search. Google brought Argon to AI Mode for Pro and Ultra subscribers. Neon's announcement doesn't mention AI Mode.
- Cyber defenders. Through Google's Fairwind Program, approved defenders can apply for a version of Argon without cyber guardrails. For Neon, Google says eligible Fairwind members start getting access over the next few days.
On OpenRouter, Neon is listed through Google AI Studio and Google Vertex, at the same $3 and $15. OpenRouter routes Neon to a European Vertex endpoint for customers who need EU data handling, and says Argon is not available that way.
Which one to pick
Pick Neon for new work. It scores higher on every shared test Google published, finishes index tasks in half the time on Artificial Analysis, and from December 1 it is the cheaper of the two. If your app already calls gemini-4-argon, test Neon before November 30, because your bill for the same traffic doubles on December 1 unless you switch.
Argon still makes sense in two cases. Until November 30, it is the cheaper per token, so a batch job that has to finish this month and doesn't need Neon's quality costs a third less on Argon. And if your work matches the science and medical tests Google ran only on Argon, you have published numbers for Argon and none yet for Neon.
If price matters more than quality, look past both. Gemini 4 Hydrogen costs $1.50 and $7.50 and scores about the same as Argon on Artificial Analysis, as our Hydrogen vs Argon comparison shows. For the top of the family, see Gemini 4 Carbon vs Argon.
The date to watch is December 1, when Argon's listed price of $4 and $20 takes effect. After that, Argon costs more than Neon on every tier of the pricing page.
Frequently asked questions
Is Gemini 4 Neon cheaper than Argon?
Not until December 1, 2026. Neon costs $3 and $15 per million input and output tokens, against $2 and $10 for Argon through November 30. From December 1 Argon rises to $4 and $20, so Neon becomes 25% cheaper.
Is Gemini 4 Neon free?
Not in the API: Google's pricing page says Neon is not available on the free tier. In the Gemini app it is offered to Google AI Pro and Ultra subscribers, not free users.
What is the Gemini 4 Neon context window?
1 million tokens, shared between input and output, the same as Argon. A normal request returns up to 64K tokens, and up to 1M through background long decode.
Is Gemini 4 Neon better than Gemini 4 Argon?
On the published numbers, yes. Neon scores 10 points higher on the Artificial Analysis Intelligence Index at the high setting and leads on all eight benchmarks Google published for both models.
Sources
- Introducing Gemini 4 NeonGoogle
- Introducing Gemini 4 ArgonGoogle
- Gemini API pricingGoogle
- Gemini API modelsGoogle
- Gemini 4 Neon model cardGoogle DeepMind
- Gemini 4 Carbon model cardGoogle DeepMind
- Gemini 4 Argon model cardGoogle DeepMind
- Gemini app release notesGoogle
- Gemini 4 Neon: intelligence, performance and price analysisArtificial Analysis
- Gemini 4 Argon: intelligence, performance and price analysisArtificial Analysis
- LLM leaderboardArtificial Analysis
- Gemini 4 Neon on OpenRouterOpenRouter
- Gemini 4 Argon on OpenRouterOpenRouter




