
Gemini 4 vs Gemini 4 Neon: Four Models Answer to One Name, and the Newest One Is Now the Default
- OrcaNEWOrcaCyber: Hy4 Flash Abliterated / Uncensored2026-11-18$0.07 / $0.28 per 1M tokens · 184 tok/s
- tencentNEWTencent: Hy4 Flash2026-11-16$0.02 / $0.07 per 1M tokens · 223 tok/s
- GoogleNEWGoogle: Gemini 4 Carbon2026-11-10$2.00 / $10.00 per 1M tokens · 87 tok/s
- AnthropicNEWAnthropic: Claude Fable 5.22026-11-09$8.00 / $40.00 per 1M tokens · 96 tok/s
- orcaNEWOrca: OrcaC2 1.02026-11-079 tok/s
- OrcaOrcaCyber: GLM 5.3 Abliterated / Uncensored2026-11-04$3.00 / $5.00 per 1M tokens · 90 tok/s
- OrcaOrcaCyber: Hy4 Abliterated / Uncensored2026-11-04$0.80 / $1.60 per 1M tokens · 131 tok/s
- OrcaOrcaCyber: Kimi K3.1 Pro Abliterated / Uncensored2026-11-04$2.00 / $5.00 per 1M tokens · 77 tok/s
- tencentTencent: Hy42026-11-04$0.83 / $2.50 per 1M tokens · 94 tok/s
- OrcaOrcaCyber: MiniMax M3.1 Flash Abliterated / Uncensored2026-10-29$0.80 / $1.60 per 1M tokens · 114 tok/s
- OrcaOrcaCyber: GLM 5.3 Flash Abliterated / Uncensored2026-10-27$0.80 / $1.60 per 1M tokens · 69 tok/s
- OrcaOrcaCyber: Qwen3.8 Flash Next Abliterated / Uncensored2026-10-26$0.80 / $1.60 per 1M tokens · 94 tok/s
- OrcaOrcaCyber: MiMo V2.6 Flash Abliterated / Uncensored2026-10-26$0.80 / $1.60 per 1M tokens · 41 tok/s
- MoonshotMoonshotAI: Kimi K3.1 Pro2026-10-2447Intelligence
- OrcaOrcaCyber: DeepSeek V4.1 Flash Abliterated / Uncensored2026-10-23$0.80 / $1.60 per 1M tokens · 209 tok/s
- minimaxMiniMax: M3.1 Flash2026-10-1929Intelligence
- OrcaOrca: OrcaCyber Ultra 1.02026-10-17$5.00 / $10.00 per 1M tokens · 111 tok/s
- deepseekDeepSeek: DeepSeek V4.1 Flash 10172026-10-1733Intelligence
- DeepSeekDeepSeek: DeepSeek V4.1 Flash 09102026-10-17$0.15 / $0.60 per 1M tokens · 311 tok/s
- DeepSeekDeepSeek: DeepSeek V4.1 Flash2026-10-1740Intelligence
Search "Gemini 4" and you will not find a model. You will find a heading — the heading Google puts above four separate models in its API reference, each with its own identifier, its own rate card and its own thinking ladder. As of 18 November 2026 that list is Gemini 4 Carbon, Gemini 4 Neon, Gemini 4 Hydrogen and Gemini 4 Argon, and there is no gemini-4 endpoint, no gemini-4 price, and no single thing the phrase refers to. Gemini 4 Neon is the one that made this worse rather than better: it is the fourth claimant, it launched last, and it is the model Google will now hand you by default in its own app.
The practical version of the question is not "which is better." It is "when somebody says Gemini 4, which of the four do they mean, and does my code care?" The answer differs by surface, and the surfaces disagree with each other. That disagreement is the thing worth writing down.
The string, resolved four ways
Google's documentation is unambiguous once you read the model reference rather than the marketing: four entries under the Gemini 4 heading, four model codes, four prices. What is less obvious is that every surface that says "Gemini 4" resolves the ambiguity differently, and none of them resolve it to a model actually called Gemini 4.
• Gemini API — no gemini-4 identifier exists. You must name one of gemini-4-carbon, gemini-4-neon, gemini-4-hydrogen or gemini-4-argon, or the call fails. This is the one surface where the ambiguity is impossible to fall into by accident.
• Google AI Studio — four model cards, listed Carbon, Neon, Hydrogen, Argon, with Neon second and described as "frontier-level intelligence with fast end-to-end task completion." All four share the same context, max output and modality line, so the card is not going to choose for you either.
• The Gemini app — a model dropdown, and for Pro and Ultra subscribers Neon is now in it. This is where "Gemini 4" gets fuzzy for the largest number of people, because the app shows a name, not a rate card.
• Google Antigravity — the harness decides. Google describes one pattern explicitly: a Neon orchestrator directing parallel Hydrogen subagents. If you use /swarm, the model answering your prompt and the models doing the work may not be the same model, and neither may be the one you selected.
So the four-way split is not a documentation problem you can resolve by reading harder. It is a real fork: the phrase means a generation in the docs, a dropdown option in the app, a routing decision in Antigravity, and nothing at all in the API.
Where Neon actually differs, since the spec sheet says it doesn't
Start with what is identical, because four models sharing a spec sheet is unusual and it changes what "vs" can even mean here. Every Gemini 4 model — Carbon, Neon, Hydrogen, Argon — runs a unified 1M-token context, decodes up to 64K output tokens by default with up to 1M available, accepts text, image, audio, video and PDF, returns text, supports native tool use and structured output, and offers implicit caching only. On paper the four are one model with four names.
What separates Neon from the other three is small in count and large in effect:
• Thinking ladder — Neon: low, medium, high, xhigh, default high. Same as Hydrogen. Carbon and Argon stop at high and default to medium.
• Price — Neon $3.00 / $15.00 / $0.15 cached, flat with no expiry. Hydrogen $1.50 / $7.50 / $0.075 and free on the free tier. Carbon and Argon $2.00 / $10.00 / $0.10 through 30 November, then $4.00 / $20.00 / $0.20.
• Free tier — Hydrogen only. Neon, Carbon and Argon have none.
• Cyber access — Neon and Carbon both have a Fairwind Program variant. Carbon's opened to everyone on 10 November; Neon's cyber variant is still gated to trusted defenders as of 19 November.
• Stated role — Carbon for the hardest long-horizon projects, Hydrogen for fast everyday execution, Argon for faster responses and lower token spend per turn, Neon for "frontier-level intelligence with fast end-to-end task completion."
The one line that matters if you are choosing blind: Neon is the only model in the family that combines the xhigh ceiling with a price that does not change in December. Hydrogen has the same ladder at half the rate, but Hydrogen's ceiling is a different model's ceiling — Google positions it for everyday execution, not frontier reasoning. Carbon has more depth, and a bill that doubles in twelve days.

The ordering question people actually get wrong
The intuitive reading of four models in a family is a ladder: cheap at the bottom, strong at the top, each one a strict improvement on the last. Gemini 4 is not shaped that way, and Neon is the release that makes the shape visible.
On Google's own numbers, Neon at xhigh scores 61 on the Artificial Analysis Intelligence Index and ranks third overall; Carbon at launch scored 59 on index v5.0 and ranked first at the time, and Artificial Analysis's own page now lists Carbon at 60 at high, seventh of 92. Argon sits below both. Hydrogen is positioned below all of them on depth and above all of them on speed. That is not a ladder, it is a portfolio with three overlapping middles.
Which means "Gemini 4 vs Gemini 4 Neon" has a different answer depending on which of the other three you had in mind:
• Against Carbon — Neon is faster and cheaper per completed task by Google's account, Carbon is the one Google points at for the hardest problems, and the price order flips on 1 December.
• Against Hydrogen — Neon buys you the same thinking ladder and the same 1M context at twice the price, with no free tier. The difference is positioning, not specification, and it is the hardest of the three comparisons to justify from a spec sheet alone.
• Against Argon — Argon is the older efficient model at the introductory rate with a December step; Neon is the newer efficient model at a flat rate with a higher ceiling. Argon has no xhigh.
If you take one rule away: do not size Neon against Hydrogen by reading the specs, because the specs are equal. Size it against Hydrogen by running your own task at high on both and comparing tokens to finish. That is the only axis on which the two differ, and it is the axis Google's per-task claims are about.
What nobody has independently measured yet
Two of Neon's most quotable numbers are less solid than they look, and both are worth flagging before they end up in a slide.
The first is the index score. Artificial Analysis had no Gemini 4 Neon model page when we checked on 19 November — its model index lists Carbon and Hydrogen from this family and nothing else. The 61 and the third place are Google citing Artificial Analysis, not Artificial Analysis publishing a Neon entry anyone can inspect. That is normal for a launch that is one day old, and it will resolve; until it does, the figure should be attributed rather than repeated as settled.
The second is the per-task cost claim. Google says Neon at default high costs 32% less per task than Carbon, and 36% or more less than competing frontier models at their highest thinking setting. Per-task cost is a function of how many tokens a model emits, and Neon's verbosity is not published anywhere in the announcement. For contrast, Artificial Analysis measures Carbon at 205M output tokens across the Intelligence Index — very verbose against a 65.1M median — at $7.76 per task and 36.0 minutes per task. The direction of Google's claim is credible precisely because Carbon's profile is that heavy. The magnitude is unaudited.
Everything else in the announcement splits cleanly into two piles. Third-party leaderboards you can go and read: Vals Index 70.9% and rank #2, a debut at #2 on Arena's Agent Arena Index with 1336 across 4,200+ sessions, first on Zapier AutomationBench at 56.3%, first on Vending-Bench 2 at $16,843, Agents' Last Exam v1.1 at 75.8%. And Google's own reported evaluations: Chartography 73.5%, Terminal-Bench Science 70.5%, RiemannBench 84.0%, HLE-Diamond 71.3%, Terminal-Bench 4.1 at 65.2%, FrontierCode 2.0 at 46.2%, the Artificial Analysis Coding Agent Index at 58.6. The second pile is not less interesting — it is just not reproducible by you.
Calling it today
In the API, the way to stop the naming problem from costing you money is to never write "Gemini 4" into a config file. Name the identifier. On our side that means google/gemini-4-carbon or google/gemini-4-argon, both available now, both at provider list price with no markup, both carrying the same December step Google published — Carbon and Argon go from $2.00 and $10.00 to $4.00 and $20.00 per million tokens on 1 December, and that change shows up in your bill the day it shows up in Google's, because there is no markup layer to absorb or delay it.
Gemini 4 Neon is not in our catalogue yet, and the honest reason to say so is that a rate card is not an endpoint. When it lands it will land on the same terms, and the thing that will make it usable in production on day one is not the price but the failover chain underneath it: a request that names Neon and falls back to Carbon if Neon is rate-limited or errors is a request that does not care which of the four Google shipped this month.

How to answer "which Gemini 4" in one pass
If the person asking is a developer writing code — the answer is that there is no Gemini 4, and they should be naming gemini-4-neon for high-frequency tool-heavy agent work, gemini-4-carbon for hard long-horizon problems, or gemini-4-hydrogen for everyday execution at volume. If they are a Pro or Ultra subscriber asking what the new option in the app is — it is Neon, and it is the same model the API sells at $3 and $15, tuned by Google's app rather than by you. If they are running Antigravity — the model that answers and the models that work may be different, and the orchestrator is likely Neon with Hydrogen subagents behind it.
The question that is still genuinely open is whether Neon's flat $3/$15 survives the arrival of Carbon's $4/$20. A model that is 50% more expensive per token than the flagship on launch day, and 25% cheaper thirteen days later without changing its own price, has not been tested by the market yet. Watch what happens to Carbon's volume in December. If it holds, Google has priced Neon as an orchestrator and Carbon as a specialist, and the four-way family makes sense. If traffic moves, the family has one default and three alternatives, and the answer to "Gemini 4" becomes Neon by acclamation rather than by documentation.

