Gemini 4 Neon: Google's $3 Middle Tier, Priced Above Carbon Until December
Google released Gemini 4 Neon on Nov 18, 2026 at $3/$15 per 1M, flat to 1M. Where it sits between Hydrogen and Carbon on price, who should switch, and official scores pending.

Key Numbers
Gemini 4 Neon · Nov 18, 2026
standard paid tier
includes thinking tokens
unified input, thinking, output
input, output, and cache
from December 1, 2026 ($4 / $20)
Flat to 1M tokens. Paid tier only. Default thinking: high.
Generally available Nov 18, 2026 on the Gemini API paid tier, Vertex AI for verified enterprise customers, and OpenRouter. No free tier. Official scores pending.
At a Glance
Gemini 4 Neon is Google's new middle tier: $3 input and $15 output per million tokens, flat across a 1M window, with no free tier and no official scores yet. Through November 30, 2026 it costs more per token than Carbon. From December 1, 2026 it sits cleanly between Hydrogen and Carbon. The case for switching rests on speed and cost per task (fewer output tokens than Carbon, by Google's account), not on proven capability.
- Generally available November 18, 2026. Stable API id:
gemini-4-neon. Model page: Gemini 4 Neon. - Price: $3 input / $15 output per million tokens, flat across a 1,048,576-token unified window. Cached input $0.15. Paid tier only.
- Twice Gemini 4 Hydrogen ($1.50 / $7.50). 25% under Gemini 4 Carbon's list from December 1, 2026 ($4 / $20). 50% above Carbon's introductory $2 / $10 through November 30, 2026.
- No free tier. No introductory price for Neon.
- Default thinking is
high;xhighis available (same shape as Hydrogen). - Google has not published official Neon benchmark scores. Capability claims below are Google's descriptions.
- Google names Neon the upgrade for Gemini 3.1 Pro Preview, with the clearest savings on prompts over 200K tokens.
What Gemini 4 Neon Is
Google describes Gemini 4 Neon as frontier-level intelligence with fast end-to-end task completion across complex software engineering, knowledge work, and science workflows. In Google's words, it sits between Gemini 4 Hydrogen and Gemini 4 Carbon for high-frequency complex tasks: deeper reasoning than Hydrogen and faster task completion than Carbon. Google says it finishes faster with fewer output tokens than Carbon while handling harder reasoning than Hydrogen, and that you should choose it where Hydrogen is not enough and you do not need Carbon's extra verification.
Google's speed labels put Carbon at Deliberate, Neon at Moderate, Hydrogen at Fast, and Gemini 3.5 Flash-Lite at Fastest. Surfaces: Gemini API paid tier only (generateContent and the Interactions API); Google AI Studio on Pro and Ultra plans; Vertex AI (GA for verified enterprise customers in select regions: global, US, EU); the Antigravity agent in Gemini Managed Agents (alone or paired with Hydrogen); and OpenRouter as google/gemini-4-neon. Knowledge cutoff is March 2026. See Google's what's new notes, model page, and changelog.
Specs
Context is a unified 1,048,576-token window shared by input, thinking, and output. Output defaults to 65,536 tokens per synchronous request (also the single-request cap, streaming or not). Long decode reaches up to 1,048,576 minus input with background: true and max_output_tokens above 65,536, or with client-side continuation in the latest SDKs. If you set background without raising max output, it still stops at 65,536.
Input: text, image, video, audio, PDF (Office and OpenDocument files also accepted). Output: text only. No image generation, no audio generation, no Live API. Tools include code execution, computer use, file search, function calling, Google Search and Maps grounding, structured outputs, and URL context. Consumption modes: Standard, Batch, Flex, Priority. Caching is implicit only on prompts of at least 4,096 tokens (95% off cached input). Explicit caching is not supported. Sampling parameters (temperature, top_p, top_k) are ignored; out-of-range values return a 400. Google says hard prompts can take several minutes, so use background execution or client timeouts of at least 15 minutes.
What Google claims Neon does
According to Google, Neon does repository-level debugging and multi-file refactoring in fewer steps with clean edits; in long tool-heavy sessions it reduces redundant file reads, handles context compaction reliably, and minimizes fabricated claims and task shortcuts; it synthesizes dense documents, spreadsheets, and charts across the 1M window; it writes code with fewer security vulnerabilities and reduces agentic overreach; and its dual-use offensive cyber capabilities are offered to defenders through Google's Fairwind Program. Those are Google's descriptions. None of them come with published Neon scores.
Where Neon Sits
The figure below places the Gemini 4 family by output price and by Google's stated reasoning order. Vertical position is ordinal only. It is not a measured capability axis.
Family position · price vs Google's ladder
output $ per 1M
Google's positioning: reasoning depth
as described by Google
Dec 1Hydrogen
$7.50
high, xhigh
Fast
Neon
$15
high, xhigh
Moderate
Carbon
$10 · to Nov 30
medium, max high
Deliberate
Carbon
$20 · from Dec 1
medium, max high
Deliberate
price / 1M out
The ladder flips on December 1
Through November 30, 2026 the paid-tier input ladder is Hydrogen $1.50, Carbon $2, Neon $3. From December 1, 2026 it becomes Hydrogen $1.50, Neon $3, Carbon $4. Neon has no introductory price, so it was priced for the post-promo ladder. Through November 30, Carbon is cheaper per token than Neon. Say that plainly when you plan a switch. If this post is still open after November 30, the intro window is closed and the Dec 1 order is the live ladder.
That flip is the main reason Neon looks expensive today and mid-tier later. Google released Carbon on the Gemini API paid tier on November 10, 2026 with a short introductory price. Neon arrived eight days later at its permanent list. The middle seat only exists cleanly once Carbon moves to $4 / $20.
Neon is a token bet
Output price includes thinking, so cost per task depends on how many tokens each model spends. At Carbon's December 1 list price, Neon matches Carbon's cost per task even if it spends 1.33x Carbon's tokens. At Carbon's introductory price, Neon only wins if it spends under 0.67x. Google gives no token-use numbers, so neither bar is proven. Google's claim that Neon uses fewer output tokens than Carbon is the entire value case until scores land.
Batch is a quiet anchor here. Neon on Batch or Flex ($1.50 / $7.50) costs exactly Hydrogen's standard price. If you can tolerate Batch or Flex latency, Neon's middle-tier intelligence (as Google describes it) lands at the same per-token rate as Hydrogen Standard. Priority goes the other way: Neon Priority at $5.40 / $27 sits 25% under Carbon Priority from December 1 ($7.20 / $36).
Neon thinks like Hydrogen, not Carbon
Neon defaults to high and offers xhigh, matching Hydrogen. Carbon (and Argon) defaults to medium and caps at high. A Carbon user who switches without setting thinking_level is running a heavier default. That changes both latency and bill, before any quality delta shows up.
Carbon, Neon, and Hydrogen share a request schema, window, and tools, so you can move steps between them with previous_interaction_id. Google's own pitch is that kind of pairing: use Neon where Hydrogen is not enough, and keep Carbon for turns that need extra verification.
Benchmarks
Google has not published official Gemini 4 Neon benchmark scores. The block below holds the three same-method rows Google already published for Hydrogen and Carbon, with Neon left empty so the real numbers can drop in later. Hydrogen and Carbon values are self-reported by Google and are not independently verified.
Official scores pending
Google has not published benchmark results for Gemini 4 Neon yet. We will add its official scores here when they are released. Until then, these are the published numbers for the two models Google places on either side of it.
Bracket rows · same published methodology
OSWorld-2.1 (offline subset, partial score)
Vending-Bench 2 (final balance)
Chartography (no tools)
When scores land, the questions that matter are narrow. Does Neon sit closer to Carbon or Hydrogen on these bracket rows? And how many output tokens does Neon spend per task versus Carbon? That second number decides whether the middle-tier price holds up. Until then, do not interpolate between the siblings or treat Google's positioning copy as a measured ranking.
The three rows above are the only same-method family comparisons this post uses: OSWorld-2.1 (offline subset, partial score), Vending-Bench 2 (final balance), and Chartography (no tools). They already appear in our Hydrogen post. Neon is pending on every row. When Google publishes Neon results, those cells will be filled and this section will be updated.
Pricing
Gemini API paid tier, flat across the full 1M window (no long-context surcharge). Every token is billed, including thinking tokens. Google: Neon has no introductory price. That matters for planning: there is no temporary Neon discount to race toward or wait out. The only calendar price move that changes Neon's relative seat is Carbon's climb from $2 / $10 to $4 / $20 on December 1, 2026. Full list: Gemini API pricing.
| Mode | Input | Output (incl. thinking) | Cached input |
|---|---|---|---|
| Standard | $3.00 | $15.00 | $0.15 |
| Batch | $1.50 | $7.50 | $0.075 |
| Flex | $1.50 | $7.50 | $0.075 |
| Priority | $5.40 | $27.00 | $0.27 |
Search and Maps grounding: 5,000 free requests a month shared across Gemini 3 and newer models, then $14 per 1,000. Vertex AI global matches Standard at $3.00 / $15.00 / $0.15. Non-global endpoints are 10% higher ($3.30 / $16.50 / $0.165). See Vertex AI pricing.
Family ladder
Sibling prices on the Gemini API standard paid tier. Argon matches Carbon. Carbon has been GA on the Gemini API paid tier since November 10, 2026.
| Model | Input | Output | Cached | Free tier | Default thinking |
|---|---|---|---|---|---|
| Gemini 4 Hydrogen | $1.50 | $7.50 | $0.075 | Yes | high (has xhigh) |
| Gemini 4 Neon | $3.00 | $15.00 | $0.15 | No | high (has xhigh) |
| Gemini 4 Carbon, through Nov 30, 2026 | $2.00 | $10.00 | $0.10 | No | medium (max high) |
| Gemini 4 Carbon, from Dec 1, 2026 | $4.00 | $20.00 | $0.20 | No | medium (max high) |
| Gemini 3.1 Pro Preview, up to 200K | $2.00 | $12.00 | $0.20 | n/a | n/a |
| Gemini 3.1 Pro Preview, over 200K | $4.00 | $18.00 | $0.40 | n/a | n/a |
Batch and Flex: Hydrogen $0.75 / $3.75; Carbon $1 / $5 through November 30, then $2 / $10. Priority: Hydrogen $2.70 / $13.50; Carbon $3.60 / $18 through November 30, then $7.20 / $36. Neon Batch and Flex ($1.50 / $7.50 / $0.075) equal Hydrogen's standard price. Neon Priority ($5.40 / $27) is 25% under Carbon's Priority from December 1 ($7.20 / $36).
A fully cached 100K-token prefix costs $0.015 on Neon instead of $0.30. Vs Gemini 3.1 Pro Preview on prompts up to 200K, Neon is 50% more on input and 25% more on output. Over 200K, Neon is 25% less on input, about 17% less on output ($15 vs $18), and 62.5% less on cached input ($0.15 vs $0.40). See Google's choosing a model guide.
Thinking Levels
Neon supports four thinking levels: low, medium, high (default), and xhigh. minimal returns an error.
low: fewest tokens, lowest latency. Google says still capable on agentic tasks.medium: interactive coding and latency-sensitive tool calls.high: balanced intelligence and speed. The recommended default.xhigh: most tokens, deepest reasoning. For hard multi-step problems wherehighfalls short.
Google's advice: run evals at all four levels and pick the lowest thinking_levelthat meets your quality bar. The default is heavier than Carbon's medium, so set the level explicitly when you migrate from Carbon or Argon.
Should You Switch?
Short decision guide from Google's positioning, the price ladder, and the missing scores.
| Your situation | Recommendation | Why |
|---|---|---|
| On Gemini 3.1 Pro Preview | Move to Neon | Google's named path. Biggest savings on prompts over 200K. Re-test short-prompt costs (+50% input). |
| On Hydrogen, hitting a reasoning ceiling | Try Neon at 2x the price | Test medium first. Google places Neon above Hydrogen for harder reasoning. |
| On Carbon, before Dec 1 | Stay | Carbon is cheaper per token through November 30, 2026. |
| On Carbon, from Dec 1, high-volume tasks without hard verification needs | Test Neon | 25% cheaper per token. Google says fewer output tokens. Set thinking_level. |
| Need a free tier | Hydrogen | Neon has no free tier. |
| Choosing on quality | Wait or run your own evals | Wait for Google's scores, or eval at all four thinking levels yourself. |
| Offline or bulk jobs | Neon Batch or Flex | $1.50 / $7.50, equal to Hydrogen Standard. |
Migration checklist
From Carbon, Argon, or Hydrogen: no request schema changes. Set model to gemini-4-neon. Set thinking_level explicitly. Use a paid-tier project.
From Gemini 3.8 Flash, 3.1 Pro Preview, or earlier: change the model string. Plan for one unified 1,048,576-token window. Remove low max_output_tokens caps (for example 8,192) or raise them to 65,536, and for long outputs use background mode with 131,072 or 262,144. thinking_level defaults to high and rejects minimal. Remove explicit caching (cached_content). Remove temperature, top_p, top_k. Declare tools natively or use structured outputs (raw-text JSON tool calls with no declared tools can return MALFORMED_FUNCTION_CALL). Return thought signatures unchanged and use previous_interaction_id. Raise client timeouts to at least 15 minutes.
Google's upgrade table: Gemini 3.1 Pro Preview goes to gemini-4-neon (or gemini-4-carbon for your hardest problems). Gemini 3.8 Flash and 2.5 Pro go to Hydrogen.
FAQ
- Google released Gemini 4 Neon on November 18, 2026. The stable API id is
gemini-4-neon. It is generally available on the Gemini API paid tier, Google AI Studio (Pro and Ultra), Vertex AI for verified enterprise customers in select regions, the Antigravity agent in Gemini Managed Agents, and OpenRouter asgoogle/gemini-4-neon. - On the Gemini API paid tier: $3.00 input, $0.15 cached input (implicit), and $15.00 output per million tokens, flat across the full 1,048,576-token window. Batch and Flex are $1.50 / $7.50. Priority is $5.40 / $27.00. There is no free tier. Vertex AI global matches the standard list; non-global endpoints are 10% higher. Google says Neon has no introductory price.
- Through November 30, 2026, no: Carbon's introductory price is $2 / $10, so Neon at $3 / $15 costs 50% more per token. From December 1, 2026, yes on list price: Neon is 25% under Carbon's $4 / $20. Cost per task still depends on how many output tokens each model spends, and Google has not published those figures.
- Neon is twice Hydrogen on input, output, and cache ($3 / $15 / $0.15 vs $1.50 / $7.50 / $0.075). Both default to
highthinking and supportxhigh. Google places Neon above Hydrogen for deeper reasoning and positions Hydrogen as the faster, cheaper worker. Neon Batch and Flex ($1.50 / $7.50) equal Hydrogen's standard price. See our Gemini 4 Hydrogen launch post. Not yet. Google has not published official Gemini 4 Neon benchmark results. The bracket rows in this post show published Hydrogen and Carbon numbers (self-reported by Google, not independently verified) with Neon marked pending. We will fill those cells when Google releases the scores.
- Context is a unified 1,048,576-token window shared by input, thinking, and output. Synchronous responses return up to 65,536 tokens by default (streaming or not). Long decode extends output up to 1,048,576 tokens minus input with
background: trueandmax_output_tokensabove 65,536, or via client-side continuation. Settingbackground: truewithoutmax_output_tokensstill stops at 65,536. - Neon supports
low,medium,high(default), andxhigh. Requests withminimalreturn an error. Google's guide:lowfor fewest tokens and lowest latency;mediumfor interactive coding;highas the balanced default;xhighfor hard multi-step problems wherehighfalls short. Google says run evals at all four levels and pick the lowest that meets your quality bar. Carbon defaults tomediumand caps athigh, so a Carbon user who switches without settingthinking_levelruns a heavier default. - No. Gemini 4 Neon is paid tier only on the Gemini API. If you need a free tier, use Gemini 4 Hydrogen.
- Google names Neon as the upgrade path for Gemini 3.1 Pro Preview (or Carbon for your hardest problems). On prompts over 200K tokens, Neon is cheaper: 25% less on input, about 17% less on output ($15 vs $18), and 62.5% less on cached input. On prompts up to 200K, Neon costs 50% more on input and 25% more on output, so re-test short-prompt workloads.
- Yes. Vertex AI lists Gemini 4 Neon as generally available for verified enterprise customers in select regions (global, US, EU). OpenRouter offers it as
google/gemini-4-neonat the same price. See Google's Vertex model page.
Outlook
Google now has a full three-step Gemini 4 ladder: Hydrogen for volume and free-tier work, Neon for the middle, Carbon for deliberate verification. Neon's value depends on two things Google has not shown yet: its official scores, and its tokens per task versus Carbon. Until those arrive, treat Neon as a priced middle seat with a Google positioning story, not as a measured upgrade over Carbon.
Watch three items. Google's official Neon benchmarks and model card. Carbon's December 1, 2026 price change, which is when the ladder finally matches Neon's list. And Vertex availability widening beyond verified enterprise customers in select regions. The clearest switch that does not need new scores is off Gemini 3.1 Pro Preview on long prompts, where Neon already undercuts on list price above 200K tokens.
Continue Reading
- GPT-6.2 Astra and Sol: Fable 5.2's Price, a Third of GPT-6 Astra's Cost per Task
- Qwen4-Max: Near-Frontier Agent Work at $1.50 and $4.50
- Hy4 flash: Tencent's $0.07 Model Undercuts Haiku 5.5 and GPT-6 Luna
- Gemini 4 Hydrogen: Argon-Class Agent Work at $1.50 and $7.50, Flat to 1M
- Seedream 5.5: ByteDance's Editing Model Beats 5.0 Pro at Less Than Half the Price per Image
