
Gemini 4 Neon vs Gemini 4 Argon: Nine Points for Fifty Cents a Task
- OrcaNEWOrcaCyber: Hy4 Flash Abliterated / Uncensored2026-11-18$0.07 / $0.28 per 1M tokens · 185 tok/s
- tencentNEWTencent: Hy4 Flash2026-11-16$0.02 / $0.07 per 1M tokens · 826 tok/s
- AlibabaNEWQwen: Qwen4 Max2026-11-16$1.50 / $4.50 per 1M tokens · 407 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 · 150 tok/s
- OrcaOrcaCyber: GLM 5.3 Abliterated / Uncensored2026-11-04$3.00 / $5.00 per 1M tokens · 99 tok/s
- OrcaOrcaCyber: Hy4 Abliterated / Uncensored2026-11-04$0.80 / $1.60 per 1M tokens · 134 tok/s
- OrcaOrcaCyber: Kimi K3.1 Pro Abliterated / Uncensored2026-11-04$2.00 / $5.00 per 1M tokens · 83 tok/s
- tencentTencent: Hy42026-11-04$0.83 / $2.50 per 1M tokens · 120 tok/s
- OrcaOrcaCyber: MiniMax M3.1 Flash Abliterated / Uncensored2026-10-29$0.80 / $1.60 per 1M tokens · 108 tok/s
- OrcaOrcaCyber: GLM 5.3 Flash Abliterated / Uncensored2026-10-27$0.80 / $1.60 per 1M tokens · 67 tok/s
- OrcaOrcaCyber: Qwen3.8 Flash Next Abliterated / Uncensored2026-10-26$0.80 / $1.60 per 1M tokens · 90 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 · 212 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 · 160 tok/s
- DeepSeekDeepSeek: DeepSeek V4.1 Flash2026-10-1740Intelligence
Gemini 4 Neon and Gemini 4 Argon are the two mid-priced Gemini 4 models, and on paper they are one tier apart in a way that makes the choice look like a budget question. It is closer to the reverse. Google's independent scores put Gemini 4 Neon at 59.6 on the Artificial Analysis Intelligence Index at its default high setting, and Gemini 4 Argon at 50.4 at its own ceiling — a 9.2-point gap that no prompt engineering closes. Argon's counter-argument is real and measurable too: it is cheaper per token, and it emits far fewer of them, so it finishes an Index task for less money today. Then 30 November arrives, Argon's introductory rate doubles, and the cheaper model stops being cheaper. This is the comparison where the calendar matters as much as the benchmark.
The two rate cards, and the date they diverge
Both are paid-tier only, both carry the same 1,048,576-token unified context, both accept text, image, video, audio and PDF and emit text, and both decode up to 64K tokens synchronously with a long-decode path to a million. The specs are so close that the whole decision sits in three numbers and one date.
• Input and output price — Gemini 4 Neon $3.00 and $15.00 flat, no expiry. Gemini 4 Argon $2.00 and $10.00 through 30 November 2026, then $4.00 and $20.00.
• Cached input — Neon $0.15 per million (95% off), Argon $0.10 today and $0.20 from 1 December.
• Thinking levels — Neon defaults to high and supports low, medium, high and xhigh. Argon defaults to medium and stops at high.
• Measured intelligence — Neon 59.6 at high, 61.4 at xhigh. Argon 50.4 at high, its highest rung.
• Measured speed — Neon 110 output tokens per second, Argon 68. Time to first token is 56.5 seconds against 46.0.
• Measured cost per Index task — Neon $6.44, Argon $4.97.
The last line is the honest case for Argon and it is a good one: 23% less money per completed task, measured on the same nine evaluations by the same tracker. The line above it is the honest case against: nine points of index, on a scale where the median model in the same comparison class scores 44.
Why Argon is cheap per task and Neon is not
Cost per task is a token count wearing a price tag, so the interesting numbers are the counts. Across the Intelligence Index evaluation set, Artificial Analysis recorded 113 million output tokens for Gemini 4 Argon and 283 million for Gemini 4 Neon, on roughly similar input volumes of 7.55 and 8.12 billion tokens. Argon's entire product claim is exactly this: Google describes it as delivering Gemini 4 intelligence with lower token spend per turn than Gemini 4 Carbon, and the measurement agrees — Argon is one of the leaner reasoning models on that board.
So the shape of the trade is unusual. Neon is the model that thinks harder and writes more, and Argon is the model that says less. If your work is a bounded transformation where a wrong answer is cheap to detect and expensive to redo, Argon's lower spend per attempt is worth more than its lower ceiling. If your work is a long agent session where one wrong architectural decision poisons the next forty tool calls, nine index points is not a small premium.
There is a second-order effect that the per-task figure hides: Neon's default is high, and Google's documentation is explicit that xhigh is where much of the model's published capability is measured and that stepping down is a real change in behaviour, not a rounding error. Argon has no xhigh rung at all. When you compare "the best each model can do", you are comparing Neon at xhigh against Argon at high, which is a 11-point gap and a more expensive per-task bill for Neon than the $6.44 quoted above.


What Argon is actually for
Argon shipped on 30 September 2026 as the efficient frontier model — the one that gets you Gemini 4-level reasoning with faster responses and lower spend than Carbon across coding, agentic, math and science tasks. In the eight weeks since, the family has filled in around it: Carbon as the maximum-capability model, Hydrogen as the free-tier workhorse, and now Neon as the high-frequency frontier model that sits between Hydrogen and Carbon.
That leaves Argon in an awkward position that has nothing to do with its quality. It is the middle of a middle that now has four models in it, and it is the only one whose price is scheduled to double. The clean read is that Argon is the model for teams that have already decided the answer quality bar is "good enough" and are optimising the bill — batch classification with verification, structured extraction over long documents, agent subtasks where a cheaper planner is fine because the worker is Hydrogen.
Where Argon still wins outright is latency to first token: 46 seconds against Neon's 56.5. If your product blocks on that number and the task is short, a 10-second head start and a smaller bill is a coherent reason to stay, benchmarks notwithstanding.
The 1 December step, priced
Argon's per-task cost is a function of its rate card, so the straightforward projection is that a doubling of input and output price roughly doubles what a task costs: $4.97 becomes something close to $9.90. Neon's rate card does not move at all. From 1 December, the arithmetic inverts and stays inverted: the model that is nine points smarter at matched defaults is also the cheaper one per completed task, and the model that was sold as the efficient option is the more expensive of the two on both axes.
Two caveats, both worth stating rather than burying. A per-task cost measured at today's prices is not a promise about December's, because caching behaviour and thinking levels change the mix; implicit caching at 95% makes input-heavy workloads insensitive to input price. And Google has moved this date once already in the family's short history, so the projection is arithmetic on a published schedule, not a forecast.
What is not uncertain is the direction. If you are choosing today between two models and one of them has a price step 10 days out, the comparison you should run is not Neon versus Argon. It is Neon versus what Argon will cost you in January.
Running both instead of choosing
The structure that survives the price step is the one Google itself describes for this family: a capable model doing the planning and a cheap one doing the volume. Argon's low token spend makes it a reasonable worker; Neon's xhigh ceiling makes it a reasonable planner; Hydrogen at $1.50 and $7.50, with a free tier, is cheaper than both for pure execution.
That is where a routing layer is worth more than a benchmark table. Both Gemini 4 Argon and Gemini 4 Carbon are in our catalogue today, alongside gemini-3.1-pro-preview, and the price you see is the provider's list price passed through with no markup — which matters here specifically, because when Argon's rate card doubles on 1 December the number in your bill changes the same day it changes on Google's page, with no second contract to renegotiate. Automatic failover is the other half: a planner that times out at 56 seconds to first token can hand the same request to Argon without your application code knowing a fallback happened. Gemini 4 Neon is not in our catalogue yet, and we will not pretend a rate card is an endpoint — the honest version is that Argon is callable here now and Neon is not.

What would change this answer
Three checkable things, in order of how much they would move the decision. If Neon's score at high turns out to be a floor rather than a plateau — that is, if independent runs at medium land near Argon's 50 rather than well above it — then the nine-point gap is really a gap at one expensive setting and the per-task economics start to favour Argon even after December. If Argon's price step is deferred or softened, the case for it gets meaningfully better. And if Google ships explicit caching or a cheaper tier for Neon, the cost-per-task comparison moves by a larger margin than any benchmark revision will.
Until one of those lands, the recommendation is narrow and unglamorous. Use Argon where the task is short, verifiable and high-volume, and where 46 seconds to first token beats a better answer. Use Neon where the task is long, the failure is expensive and the extra tokens are the point. And if your volume is large enough that per-task cost decides it, put the migration behind a router so the 1 December step is a routing change rather than a rewrite.
