AI/News

Gemini 4 Neon Is Out: Price, Specs and How It Compares

Google's new mid-tier Gemini 4 model undercuts GPT-6.2 Astra on price while claiming frontier-level reasoning for coding and agentic work.

Gemini 4 Neon announcement header graphic from Google
Image: Google.

Google released Gemini 4 Neon on November 18, 2026, slotting a new mid-tier model into the Gemini 4 lineup between the heavyweight Carbon and the fast, everyday Hydrogen. Neon is priced at $3 per million input tokens and $15 per million output tokens, undercutting GPT-6.2 Astra while claiming frontier-level reasoning on coding and agentic tasks. It's rolling out now in the Gemini API, Google AI Studio, the Gemini app and Google Antigravity for paid subscribers.

Gemini 4 Neon at a glance

  • Released: November 18, 2026
  • Price: $3 / million input tokens, $15 / million output tokens (cached input: $0.15/million)
  • Context window: 1 million tokens
  • Thinking levels: four, from default to xhigh
  • Available in: Gemini API, AI Studio, Gemini app, Antigravity, Gemini for business

What is Gemini 4 Neon?

Gemini 4 Neon is the fourth model Google has shipped under the Gemini 4 name since mid-October, and the third in roughly a week, following Carbon's November 10 debut and Hydrogen's launch two days after that. Google describes Neon as built for "complex software engineering, knowledge work, and science workflows," positioning it as the balance point between Carbon, which Google reserves for the hardest long-horizon projects, and Hydrogen, which is tuned for fast, high-volume everyday tasks. Neon is meant to deliver frontier-level reasoning for complex, high-frequency work without Carbon's cost or latency.

In practice, that makes Neon the model Google wants developers reaching for by default: capable enough for serious coding and research work, but priced to run at the volume agentic workflows demand. The model supports four internal "thinking levels" that trade off latency for depth, with a new xhigh setting added for tasks that benefit from deeper multi-step reasoning. Google has not published exact latency figures for each level, but the general pattern established by Carbon and Hydrogen is that higher thinking levels take longer per response in exchange for more consistent results on multi-step problems.

The naming also continues a pattern that's become hard to miss: Argon, Carbon, Neon and Hydrogen are all elements or element-adjacent names, following Google's shift away from numbered point releases (like Gemini 3.5 or 3.8) toward short codenames for each Gemini 4 variant. Google hasn't explained the scheme publicly, but reporters and developers have already started using the element names as shorthand in place of version numbers, which is part of why searches comparing Neon directly against Carbon and Hydrogen have picked up since launch.

Price and context window

Neon costs $3 per million input tokens and $15 per million output tokens, with a steep 95% discount on cached input tokens, down to $0.15 per million. The context window is 1 million tokens, matching Carbon and Hydrogen and putting it on par with the largest windows OpenAI and Anthropic currently offer in their flagship models.

Google says that once introductory pricing for Carbon and Argon ends on December 1, 2026, Neon is expected to come in cheaper than Carbon on a per-token basis, reinforcing its role as the volume option in the lineup rather than a direct Carbon replacement.

How Neon's price compares

Neon lands well below GPT-6.2 Astra and modestly below Claude Opus 5.5 on both input and output pricing, while costing more than OpenAI's smaller Sol model:

ModelInput ($/M tokens)Output ($/M tokens)Context window
Gemini 4 Neon$3$151M tokens
Claude Opus 5.5$4$201M tokens (extended)
GPT-6.2 Sol$2$10400K tokens
GPT-6.2 Astra$8$40400K tokens

On a pure price basis, Neon sits between OpenAI's two November releases: more expensive than the compact Sol, but less than half the cost of the flagship Astra, per OpenAI's own API pricing page. Against Anthropic's Opus 5.5, whose rates are published on Anthropic's pricing documentation, Neon is marginally cheaper on both input and output tokens.

Benchmark claims

Google's own evaluations put Neon at #3 overall on the Artificial Analysis Intelligence Index with a score of 61 at its xhigh thinking level, and #2 on the Vals Index at 70.9%. The company highlights coding-specific results most heavily: Neon ranks #1 on the Artificial Analysis Coding Agent Index at 58.6, scores 65.2% on Terminal-Bench 4.1, and hits 46.2% on FrontierCode 2.0 at its default high setting.

Google also reports Neon finishing #1 on several narrower benchmarks, including Legal Research Bench, Tax Agent Bench, and Zapier's AutomationBench (56.3%), plus a reported $16,843 result on Vending-Bench 2. On security-flavored tests, Neon scores 90.5% on SEC-bench Pro and a 0.0% attack success rate on the Gray Swan indirect prompt injection benchmark at its highest thinking level.

These figures all come from Google's own model card and blog post rather than an independent lab, so treat them as the company's claims about its own model rather than verified third-party results. Early independent comparisons, when they appear, may land differently depending on the exact prompts and thinking level used.

A few of the benchmark names are worth unpacking. Terminal-Bench and FrontierCode both test a model's ability to work through realistic coding tasks inside a sandboxed environment rather than answering isolated questions, which is closer to how a developer would actually use an agentic coding tool. WebExploitBench and CWE-bench, where Neon scores 51.26% and 70.7% respectively, test security-specific reasoning: finding and explaining software vulnerabilities rather than just writing new code. The Gray Swan indirect prompt injection test matters for anyone deploying Neon as an autonomous agent that reads untrusted web content or documents, since a 0.0% attack success rate at the highest thinking level suggests the model resisted every injection attempt Gray Swan threw at it in that configuration — though Google doesn't say how many attempts that figure is based on, or how the lower thinking levels performed on the same test.

What developers switching models should know

For teams already running workloads on Hydrogen or Carbon, switching to Neon is a model-name change in most Gemini API integrations rather than a rework, since all three share the same 1-million-token context window and the same four-tier thinking-level system. The practical decision is almost entirely about cost and task complexity: Hydrogen for high-volume, lower-stakes requests; Neon for coding-heavy or multi-step agentic work that needs more reasoning depth than Hydrogen reliably provides; and Carbon reserved for the small share of tasks that are long-horizon enough, or high-stakes enough, to justify its higher per-token cost.

Google is pushing Antigravity, its agentic coding environment, as a primary surface for Neon, which signals where the company expects the model to see the most use: autonomous or semi-autonomous coding agents that need to plan, write, and verify code across multiple steps without constant human review at each stage. Developers can also try the model directly in Google AI Studio without writing a full integration first.

Where to try Gemini 4 Neon

Neon is live now for:

  • Developers: via the Gemini API and Google AI Studio, including the paid developer tier.
  • Gemini app subscribers: Pro and Ultra tiers can select Neon from the model dropdown.
  • Antigravity users: paid Google AI Pro and Ultra subscribers get access to Neon in Google's agentic coding environment.
  • Business customers: through Gemini for business, via the API or the model dropdown.
  • Cybersecurity defenders: eligible researchers in Google's Fairwind Program get access rolling out over the following days.

Free-tier Gemini app users aren't getting Neon directly; it's reserved for paid Pro and Ultra subscribers for now, consistent with how Google handled the Carbon and Hydrogen rollouts earlier in November.

Fairwind program and business access

Alongside the standard developer and consumer rollout, Google is extending early Neon access to eligible participants in its Fairwind Program, a track aimed at cybersecurity defenders who get preview access to new Gemini models for vulnerability research and defensive tooling before, or alongside, the general release. Access for Fairwind participants is rolling out over the days following the November 18 launch rather than all at once, which is consistent with how Google staged Fairwind access for Carbon earlier in the month.

On the business side, Gemini for business customers can reach Neon the same way they'd reach Carbon or Hydrogen: either by calling the model directly through the API with the gemini-4-neon identifier, or by selecting it from the model dropdown in supported admin consoles. Google hasn't announced enterprise-specific pricing separate from the standard API rates, so business customers pay the same $3/$15 per-million-token rate as individual developers, before any volume agreements.

Neon vs. Carbon vs. Hydrogen vs. Argon

With Neon's release, Google now has four actively-supported models under the Gemini 4 name, each aimed at a different point on the cost-versus-capability curve:

ModelReleasedBest for
Gemini 4 ArgonOct 13, 2026General-purpose, earliest Gemini 4 release
Gemini 4 CarbonNov 10, 2026Hardest, long-horizon projects
Gemini 4 HydrogenNov 12, 2026Fast, high-volume everyday tasks
Gemini 4 NeonNov 18, 2026Complex coding and agentic work at lower cost than Carbon

That's four Gemini 4 variants shipped in a little over five weeks, a faster cadence than Google used for Gemini 3. It also means developers choosing a model now have real tradeoffs to weigh rather than a single default option, which is part of why direct, same-generation comparisons like Neon against Carbon and Hydrogen have become a bigger part of how people search for and evaluate Gemini releases.

Why it matters

Neon's release keeps Google shipping model updates at a pace competitors are struggling to match, and its pricing puts clear pressure on OpenAI's higher-end Astra tier in particular. For teams already building on Gemini 4, Neon is pitched as the model to reach for when Hydrogen isn't capable enough but Carbon is overkill or too expensive to run at scale — a middle tier that didn't exist in the lineup until this week.

It also arrives at a moment when the three major labs are all shipping multiple model tiers within days of each other: OpenAI split GPT-6.2 into Astra and Sol on November 17, and Anthropic's Opus 5.5 is already being compared directly against both. For buyers, that means the "which model should I use" question increasingly has to be answered per-task rather than per-vendor, since each lab now offers a cheap, a mid-tier, and a frontier option that update on overlapping schedules. Readers who want the full breakdown of what changed between the earlier Gemini 4 releases can see how Hydrogen, Carbon and Argon compare before deciding whether Neon changes that calculus.

What's next

Google hasn't said when the next Gemini 4 variant might arrive, but its cadence this quarter — four releases since mid-October — suggests another update before year's end is plausible. Carbon and Argon's introductory pricing is set to expire December 1, 2026, which will be the next concrete date to watch for how Neon's cost compares once the current promotional pricing ends. If Google follows the same playbook it used after Carbon and Hydrogen shipped, expect head-to-head benchmark posts from third-party evaluators within the first week or two, which should give a clearer picture of how Neon's claimed coding and agentic results hold up outside Google's own test suite.

For now, the practical takeaway for anyone building on Gemini is that Neon is live, it's the cheapest of the three big labs' mid-tier coding models, and switching to it from Hydrogen or Carbon inside the Gemini API requires nothing more than changing the model identifier in an existing integration.

Frequently asked questions

How much does Gemini 4 Neon cost?

Gemini 4 Neon costs $3 per million input tokens and $15 per million output tokens, with cached input priced at $0.15 per million tokens, a 95% discount.

What is Gemini 4 Neon used for?

Google positions Neon for complex software engineering, knowledge work and science workflows, sitting between the fast Hydrogen model and the more expensive, long-horizon Carbon model.

Is Gemini 4 Neon available for free?

No. Free-tier Gemini app users don't get Neon. It's available to Gemini Pro and Ultra subscribers, developers on the Gemini API and AI Studio, and Gemini for business customers.

How does Gemini 4 Neon compare to GPT-6.2 and Claude Opus 5.5 on price?

Neon ($3/$15 per million tokens) is cheaper than both GPT-6.2 Astra ($8/$40) and Claude Opus 5.5 ($4/$20), but more expensive than GPT-6.2 Sol ($2/$10).

What's the context window on Gemini 4 Neon?

Neon has a 1-million-token context window, matching Gemini 4 Carbon and Hydrogen.

What are Gemini 4 Neon's thinking levels?

Neon supports four thinking levels, from a default setting up to a deeper xhigh mode intended for multi-step reasoning tasks.

Sources

More on Gemini 4 →Gemini 4Google AIGemini 4 NeonAI pricingGoogle Antigravity
Theo Park
Written byTheo Park

Theo Park runs the AI desk at Pandromeda. He follows model launches from the frontier labs and the open-weight community, tracks the assistants and developer tools built on them, and explains what each release changes on pricing, capability and safety. His reporting leans on primary sources: model cards, technical reports, API documentation and the companies' own announcements.

More from AI

See all