Verified · Nov 20, 2026
Independently verifiedGoogle ships Gemini 4 Neon: frontier intelligence at $3/$15 per million tokens, #3 on the Artificial Analysis Intelligence Index at its xhigh level per the launch post — rolling out since Wednesday to paid Gemini API users and Google AI Pro/Ultra subscribers
3 sourcesGoogle's launch post states: "Today, we’re introducing Gemini 4 Neon — bringing frontier-level intelligence with fast end-to-end task completion across complex software engineering, knowledge work, and science workflows." On pricing, the post states: "The model is priced at $3 per million input tokens ($0.15 per million cached input tokens, a 95% discount) and $15 per million output tokens, with a 1 million-token context window." The Decoder's rate context reads: "The model is live today through the API at $3 per million input tokens and $15 per million output tokens, above Carbon and Argon's introductory rate of $2/$10 (doubling to $4/$20 on December 1) and twice the price of Hydrogen at $1.50/$7.50." and VentureBeat's placement frame reads: "Google puts Neon between Hydrogen, its small, light, fast workhorse, and Carbon, the model it recommends for its hardest extended projects." On availability, the post states: "Gemini 4 Neon is rolling out today in the Gemini API in Google AI Studio, Gemini for business (via API and model dropdown), for paid Google AI Pro and Ultra subscribers in Google Antigravity, and for Pro and Ultra subscribers in the Gemini app." On benchmarks, the post states: "Neon ranks ahead of comparable models and #3 overall with a score of 61 on the Artificial Analysis Intelligence Index at its xhigh thinking level, landing just behind larger, slower models." On coding, the post states: "Across core developer and researcher workflows, Neon achieves frontier performance and is competitive with the leading models in its class, ranking #1 overall on the Artificial Analysis Coding Agent Index (58.6), with a score of 65.2% on Terminal-Bench 4.1 and 46.2% (at its default high setting) on FrontierCode 2.0." On cost per task, the post states: "At its default high thinking level, it pairs a high Artificial Analysis Intelligence Index score with a 32% lower cost per task than Carbon." and VentureBeat's cross-model reading states: "More impressively, Neon costs 25% below Anthropic's larger, flagship Opus 5.5’s standard rates of $4 and $20 despite scoring within about a percentage point on that larger model on two engineering benchmarks." — with its own caveat: "Reasoning settings apply to the Artificial Analysis column. Providers define those settings separately, so two models running at maximum effort can spend very different amounts of time and money on a question."
Why now
The launch landed Wednesday, November 18, 2026 — The Decoder's count: "It's also Google's fourth new Gemini 4 model in four weeks." — and it lands inside a live pricing window: VentureBeat's calendar reads "Carbon’s introductory rate makes it cheaper per token than Neon through November." and "Beginning December 1, Neon’s input and output rates are 25% below Carbon’s scheduled rates." — so the Neon-versus-Carbon price comparison a creator makes this week changes on December 1. For creators on a paid Google AI Pro or Ultra plan, the post's availability sentence puts Neon in the Gemini app and Antigravity model dropdowns, and the API side is the paid tier — checkable this week in the product. And VentureBeat frames the competitive stakes: "The timing also gives Google an opening with developers weighing OpenAI’s new GPT-6.2 Astra and Sol."
Why it is worth publishing
Model-launch cards travel beyond the niche when the numbers are concrete and the tiers are checkable: this one carries headline pricing ($3/$15 with the 95% cached-input discount), a model dropdown a paid subscriber can open today in their own Gemini app, and a December 1 rate deadline that gives the why-now teeth. The risk cards keep the attribution discipline — every benchmark is Google's own reading, and cross-vendor comparisons inherit VentureBeat's reasoning-settings caveat — so the card stays explainable without turning vendor numbers into independent verdicts.
Evidence basis
Three opened sources, one official: Google's launch post (direct curl, HTTP 200, 401,328 bytes on disk; article:published_time 2026-11-18T21:00:00+00:00 with a dateModified of 2026-11-19T14:03:59.958636+00:00 recorded as the post-publication edit; November 18 = Wednesday, calendar-verified this run), The Decoder (raw HTML, 117,341 bytes on disk, article:published_time 2026-11-18T21:03:23+00:00, author Matthias Bastian), and VentureBeat (jina capture, 25,651 bytes, after direct curl HTTP 429; its Published Time agreeing to the second with the Google News RSS item's pubDate). Heat rides the cadence — Google's fourth Gemini 4 model in four weeks per The Decoder's body — the headline pricing between Hydrogen and Carbon, and the December 1 Carbon-rate deadline, tempered by every benchmark being Google's own reading under its own scoping, tier-gated availability, and the cyber version staying Fairwind-gated.
“Google's new Gemini 4 Neon is rolling out at $3 and $15 per million tokens — for paid AI Pro and Ultra subscribers in the Gemini app — and Carbon's introductory price undercuts it only through November.”
Angle
A price-ladder walkthrough built on opened sources: Hydrogen at $1.50/$7.50, Carbon's introductory $2/$10 (cheaper per token than Neon through November per VentureBeat), Neon at $3/$15, and Carbon's scheduled jump to $4/$20 on December 1 — with every benchmark carried as Google's own reading and the tier map spelled out.
Format
Long-form explainer
Demo idea
A pricing-ladder board: Hydrogen $1.50/$7.50, Carbon $2/$10 introductory (undercutting Neon through November per VentureBeat), Neon $3/$15, Carbon's December 1 jump to $4/$20 — each row labeled with its source, closing on the row a creator can verify in their own Gemini app: the model dropdown, for paid Google AI Pro and Ultra subscribers.
Platform notes
Every benchmark and cost-per-task figure is Google's own reported reading — carry the post's own scoping (“at its xhigh thinking level”, “at its default high setting”, “in internal Google Antigravity testing”) wherever a number goes, and VentureBeat's reasoning-settings caveat when comparing across vendors. The Gray Swan result is benchmark-internal at the highest thinking level — never a general “unhackable”. Availability is tier-gated: paid Gemini API tier for developers; Google AI Pro or Ultra for Antigravity and the Gemini app; the cyber version stays Fairwind-gated “in the coming days”. “rolling out today” is the announcement's voice dated November 18 — card voice says Wednesday. The model handles text, images, video, audio, and PDFs, with an about-one-million-token context window, per The Decoder — with its own boundary: "Image and audio generation aren't supported."
Usable claims
- Google's launch post states: "Today, we’re introducing Gemini 4 Neon — bringing frontier-level intelligence with fast end-to-end task completion across complex software engineering, knowledge work, and science workflows." On pricing, the post states: "The model is priced at $3 per million input tokens ($0.15 per million cached input tokens, a 95% discount) and $15 per million output tokens, with a 1 million-token context window." On availability, the post states: "Gemini 4 Neon is rolling out today in the Gemini API in Google AI Studio, Gemini for business (via API and model dropdown), for paid Google AI Pro and Ultra subscribers in Google Antigravity, and for Pro and Ultra subscribers in the Gemini app." — The Decoder's rate context reads: "The model is live today through the API at $3 per million input tokens and $15 per million output tokens, above Carbon and Argon's introductory rate of $2/$10 (doubling to $4/$20 on December 1) and twice the price of Hydrogen at $1.50/$7.50." VentureBeat adds the calendar on Carbon's rate: "Carbon’s introductory rate makes it cheaper per token than Neon through November." and "Beginning December 1, Neon’s input and output rates are 25% below Carbon’s scheduled rates." The post's own family framing states: "While Carbon tackles your hardest, long-horizon projects and Hydrogen handles fast everyday execution, Neon balances depth and efficiency." and its developer on-ramp states: "Build with the Gemini API paid tier in Google AI Studio and Android Studio, if you’re a paid Google AI Pro or Ultra subscriber, you can try Neon in Google Antigravity and all Pro and Ultra subscribers can try it in AI Studio."
- On benchmarks, the post states: "Neon ranks ahead of comparable models and #3 overall with a score of 61 on the Artificial Analysis Intelligence Index at its xhigh thinking level, landing just behind larger, slower models." On coding, the post states: "Across core developer and researcher workflows, Neon achieves frontier performance and is competitive with the leading models in its class, ranking #1 overall on the Artificial Analysis Coding Agent Index (58.6), with a score of 65.2% on Terminal-Bench 4.1 and 46.2% (at its default high setting) on FrontierCode 2.0." and "On SWE-sweep, it already leads on the mini-SWE harness (12.8%) and more than triples to a state-of-the-art 39.0% in Google Antigravity showing what the Antigravity harness unlocks." On cost per task, the post states: "At its default high thinking level, it pairs a high Artificial Analysis Intelligence Index score with a 32% lower cost per task than Carbon." and "It also sits on the Pareto frontier on both the Vals index and Agent Arena at high, with roughly 50% to 80% lower cost per task than top competing frontier models." and "Neon shows its highest intelligence score here, while still costing 36%+ less per task than competing models at their highest setting." On steerability and fabrication, the post states: "Neon ranks #2 on steerability (adapting cleanly to mid-task user corrections) on the new Agent Arena Index" and "recording our lowest fabrication rate yet in internal Google Antigravity testing" Thai Tran, AI Product Lead at Glean, quoted in the post's announcement: "In our early evaluations, it matched or exceeded Gemini 4 Carbon on overall quality at 3x the speed and 60% of the cost." VentureBeat's cross-model reading states: "More impressively, Neon costs 25% below Anthropic's larger, flagship Opus 5.5’s standard rates of $4 and $20 despite scoring within about a percentage point on that larger model on two engineering benchmarks." and adds: "Artificial Analysis confirmed on X that Neon placed third on its intelligence index." and "Reasoning settings apply to the Artificial Analysis column. Providers define those settings separately, so two models running at maximum effort can spend very different amounts of time and money on a question." and "And under OpenAI’s long-context pricing, a Sol request exceeding 272,000 input tokens costs $4 per million input tokens and $15 per million output tokens for the whole request."
- On cyber defense, the post states: "Gemini 4 Neon produces more secure code than Carbon across all easy, medium, and hard challenges on AutoBaxBench, achieving state-of-the-art on the hard tier (76%)." and "On WebExploitBench v1.2, Neon is state-of-the-art (51.26%) at half the cost of competing large models." and "Gemini 4 Neon for cyber is rolling out to trusted defenders in the coming days as part of the Fairwind Program." On the safety stack, the post states: "maintaining Gemini 4 Carbon-level protections against severe misuse and prompt injection while advancing further on model alignment and agentic behavior" — and, on real-time safeguards, "Because Neon brings frontier-level capabilities to a faster, lower-cost model, we paired it with strong real-time safeguards against cyber and biosecurity misuse comparable to our Carbon protections, while declining far less legitimate work in these areas, refusing for instance just 1.98% of tasks on Terminal-Bench 4.0 and 0% on Terminal-Bench-Science." On discipline, the post states: "It matched Carbon with 0.0% cheating on Terminal-Bench 4.0 and took shortcuts on just 2.45% of tasks on our coding cheating evaluation (the lowest rate of any Gemini 4 model)." and "It also achieved a 0.09% misaligned action rate on our agentic safety suite and a 99.2% pass rate on our agent honesty evaluation." On prompt injection, the post states: "We found that with real-time safeguards at its highest thinking level, Neon resisted every attack within this Gray Swan benchmark, matching the 0.0% attack success rate reached only by other Gemini models." Karina Nguyen, Founder of Thoughtful Lab, quoted in the post's announcement: "In our cyber evals, Neon was able to find several subtle vulnerabilities other models missed while using a fraction of the tokens."
Evidence pipeline
From the news
Breakdown
Three opened sources, one official. Google's launch post (direct curl; dated by its own published_time meta and rendered label, with a post-publication dateModified recorded) carries the announcement mechanics: the positioning sentence, the pricing sentence with its parenthetical cached-input rate, the tiered availability, the family framing, and the benchmark highs — #3 on the Artificial Analysis Intelligence Index at xhigh, #1 on the coding agent index, the 32%-lower cost per task at its default high thinking level — plus the cyber lineup (AutoBaxBench, WebExploitBench, the Fairwind-gated cyber rollout) and the discipline measurements (fabrication, cheating, misalignment, honesty, Gray Swan). The Decoder adds the family context (its body's “fourth new Gemini 4 model in four weeks”), the intro-rate math against Hydrogen and Carbon, and the capability boundary (“Image and audio generation aren't supported.”). VentureBeat adds the cross-model frame (25% below Opus 5.5's standard rates, its pricing table, the AA confirmation, the Sol long-context twist) and its reasoning-settings caveat. What stays scoped: every benchmark is Google's own reading; availability is tier-gated; the cyber version is days away and Fairwind-only; and the December 1 deadline comes from VentureBeat's calendar.
Sources
Risks
- Before you quote a number, open the launch post and carry its own scoping with the figure — “at its xhigh thinking level”, “at its default high setting”, “in internal Google Antigravity testing”. When you compare across vendors, keep VentureBeat's reasoning-settings caveat attached: two models at maximum effort can spend very different time and money. Present the Gray Swan result as a benchmark-internal finding at the highest thinking level, never as a product guarantee.
- When you tell viewers where to try Neon, name the tier in the same breath: paid Gemini API tier for developers, Google AI Pro or Ultra for Antigravity and the Gemini app. Say the cyber version stays Fairwind-gated and is still days out per the post. Anchor timing to Wednesday, November 18 — the announcement's own day — never to “today”.
Demo ideas
- A pricing-ladder board: Hydrogen $1.50/$7.50, Carbon $2/$10 introductory (undercutting Neon through November per VentureBeat), Neon $3/$15, Carbon's December 1 jump to $4/$20 — each row labeled with its source, closing on the tier a creator can verify in their own Gemini app: the model dropdown, for paid Google AI Pro and Ultra subscribers.
- A dropdown walkthrough: open the Gemini app as a Google AI Pro or Ultra subscriber, select Gemini 4 Neon, and run one real task from the post's own examples — analyzing billing statements to spot why a utility charge spiked — narrating what the model does differently from the tier you usually use.
- An “introductory rate” explainer: what “introductory” pricing means when it expires (Carbon's $2/$10 through November per VentureBeat, doubling to $4/$20 on December 1), how Neon's $3/$15 sits against it, and the check a creator runs before locking a workflow to any model's current price.