Gemini 4 Neon vs Claude Opus 5.5: Price, Benchmarks, Animation
Gemini 4 Neon, released on 18 November 2026, costs $3 per million input and $15 per million output tokens with no free tier, against $4 and $20 for Claude Opus 5.5. It scores 61 on Artificial Analysis at xhigh and ranks fifth on Arena WebDev, behind three Claude models. Checked 20 November 2026.
Rédaction assistée par IA : ce guide a été rédigé avec Claude Sonnet 5.5, et chaque fait renvoie à ses sources.
Gemini 4 Neon is Google's middle Gemini 4 model, released on 18 November 2026 between Carbon and Hydrogen (Google). It costs $3 and $15 per million tokens, paid only, and is the default model in Antigravity. It is cheaper per task than Claude Opus 5.5, but three Claude models still lead Arena WebDev.
What is Gemini 4 Neon and what does it cost?
Gemini 4 Neon (`gemini-4-neon`) went GA on 18 November 2026 at $3 per million input tokens and $15 per million output tokens, thinking included, with a 1,048,576-token context window and no free tier (pricing, checked 20 November 2026). Google's model card places it between Carbon and Hydrogen and says it is adapted from Carbon.
| Item | Gemini 4 Neon | Source |
|---|---|---|
| Model ID | `gemini-4-neon`, stable only, no preview or dated alias | Model page |
| Release | GA on 18 November 2026; no shutdown date | Changelog, deprecations |
| Standard price per 1M tokens | $3.00 input, $15.00 output, $0.15 cached input; flat across the window | Pricing |
| Batch and Flex | $1.50 input, $7.50 output | Pricing |
| Priority | $5.40 input, $27.00 output | Pricing |
| Free tier | None; thinking tokens are billed as output | Pricing |
| Agent Platform (formerly Vertex AI) | Same on the global endpoint; $3.30 and $16.50 elsewhere; a $3.75 per 1M cache-write charge from 12 January 2027 | Model Garden pricing |
| Context window | 1,048,576 tokens, shared by input, thinking and output | Model page |
| Max output | 65,536 tokens per synchronous request; up to 1,048,576 minus input with long decode | What's new |
| Thinking levels | low, medium, high (default), xhigh | What's new |
| Input and output | Text, image, video, audio and PDF in; text out | Model page |
| Knowledge cutoff | March 2026 | Model card |
A request of 3,000 input and 12,000 output tokens, about one animation file, costs $0.189 on Neon and half that on Batch, against $0.252 on Claude Opus 5.5 (Opus 5.5 docs) and, from 1 December, on Argon or Carbon. Agent loops cost more. The Gemini 4 hub guide compares all four models.
Who can use Gemini 4 Neon, and where?
Neon is paid only. Since 18 November 2026 it runs on the paid Gemini API and in AI Studio, in the Gemini app for Google AI Pro ($19.99 a month in the US) and Ultra subscribers, and in Antigravity as the default model (Gemini Apps Help, checked 20 November 2026). GitHub Copilot and Cursor offer it on paid plans.
| Surface | Neon access | Source |
|---|---|---|
| Gemini API | Paid tier only | Changelog |
| Google AI Studio | Paid API keys; AI Pro and Ultra subscribers can also try it | What's new |
| Gemini app | AI Pro and Ultra, in Chats and Tasks. US plans cost $19.99, $99.99 or $199.99 a month | Gemini Apps Help, plans |
| Antigravity app, CLI and IDE extensions | Paid AI Pro and Ultra; default since 18 November 2026. Orchestrator (Neon) plans with Neon, implements with Hydrogen. CLI API-key use since v1.5.12 (19 November) | Changelog, Antigravity post |
| Agent Platform | Verified enterprise customers, in global, us and eu | Model page |
| GitHub Copilot | Pro+, Max, Business and Enterprise; not Copilot Pro. Copilot drops Argon on 14 December 2026 | GitHub Docs, changelog, Argon notice |
| Cursor | $3 input, $0.15 cache read, $15 output per 1M; Premium, Super and Ultra plans | Cursor docs |
| OpenRouter | Community slug `google/gemini-4-neon`, served by AI Studio and Vertex | OpenRouter |
| Gemini CLI | Not official; the README says it is not tuned for Gemini 4 and points to Antigravity CLI | README |
What does Google claim for Gemini 4 Neon?
Google's launch post of 18 November 2026 ranks Neon #3 on Artificial Analysis at 61, #2 on the Vals Index at 70.9% and #2 on Agent Arena at 1,336. On 20 November the first two matched the live boards. The Agent Arena board showed Neon (High) #3 at 1,333, preliminary; 1,336 was Claude Opus 5.5 at max.
motif's research reproduced 32% only one way: Carbon at high as Artificial Analysis showed it from 12 to 18 November, $5.48 per task at the $2/$10 introductory price (archived AA page), against $3.74 for Neon at high. Against Carbon's default medium at its 1 December list price, Neon high is 43% cheaper. This is an inference, not a Google statement.
Google makes no frontend or UI quality claim for Neon. The only UI wording in its docs is that the medium thinking level suits "live UI prototyping" (what's new). Google's frontend and motion claims belong to Carbon and Hydrogen; see the Carbon guide and the Hydrogen guide.
How does Gemini 4 Neon score against Claude on independent boards?
On boards Google does not run, Neon is near the top but not first. On 20 November 2026 Artificial Analysis put Neon at 61 at xhigh, third on its model leaderboard, and Arena WebDev ranked it fifth at 1,763, behind Claude Fable 5.2, Opus 5.5, Sonnet 5.5 and Gemini 4 Carbon. Where boards compare cost with Claude, Neon is cheaper.
| Board | Gemini 4 Neon | Claude | Source |
|---|---|---|---|
| Artificial Analysis Intelligence Index v5.0.1 | 61 (xhigh), $6.44 per task; #3 on the model leaderboard, #6 of 95 when each thinking level counts | Fable 5.2 65, $26.54; Opus 5.5 59, $10.04 (both max) | AA leaderboard, AA Neon page |
| AA Coding Agent Index v2.1 | 59 in Antigravity CLI (xhigh), $6.86 per task | 58 each: Fable 5.2 (xhigh) in Claude Code, $15.47; Opus 5.5 (max), $12.61, harness not named separately | AA article |
| Arena WebDev, overall | #5, 1,763 (high) | Fable 5.2 #1 1,823; Opus 5.5 #2 1,818; Sonnet 5.5 #3 1,771 | Arena |
| Arena WebDev, HTML category | #5, 1,788 (high) | Fable 5.2 #6 1,778; Opus 5.5 #9 1,753 | Arena HTML |
| Arena Agent Arena Index | #3, 1,333 ±32 over 4,238 sessions, preliminary (high) | Fable 5.2 #1 1,448; Opus 5.5 #2 1,336 (max) | Agent Arena |
| Vals Index v2.2 | 70.87%, #2 (xhigh) | Fable 5.2 71.31%; Opus 5.5 70.14% | Vals Index |
| Vals Vibe Code Bench v1.1 (working web apps) | 93.70%, #2, $6.76 per test | Fable 5.2 93.77% #1, $22.13; Opus 5.5 90.29% #7, $35.77 | Vibe Code Bench |
| Design Arena, combined Overall | #1, 1394 on 1,239 battles | Opus 5.5 #5 1378; Fable 5.2 #7 1369 | Design Arena |
| Design Arena, 3D Design (HTML) | #2, 1502 on 349 battles | Fable 5.2 #3 1486; Opus 5.5 #5 1450 | Design Arena 3D |
Read it by task. Arena voters prefer Claude for full sites, games and simulations: in Gaming Opus 5.5 leads at 1,944 and Neon is fifth at 1,866. On single HTML files and working-app tests Neon is level or ahead, and its Design Arena lead rests on few battles. No board tests animation or code-to-video yet.
What have creators made with Gemini 4 Neon?
In Neon's first two days, creators posted Three.js scenes, pure JavaScript motion pieces and 3D games, mostly made in Antigravity at xhigh. None of the praise posts ran the same prompt on Claude. The one same-task comparison found, Peter Yang's of 19 November 2026, says Opus 5.5 still wins on 3D, games and design (video).
Gemini 4 Neon or Claude Opus 5.5: which is better for animation and web work?
It depends on the job. Neon is 25% cheaper per token than Opus 5.5 ($3/$15 vs $4/$20, checked 20 November 2026), cost $6.86 vs $12.61 per AA Coding Agent Index task, and leads on single HTML files and working-app tests. Opus 5.5 ranks higher on Arena WebDev overall and in Frontend, Gaming and Simulations.
| Gemini 4 Neon | Claude Opus 5.5 | Source | |
|---|---|---|---|
| API model ID | `gemini-4-neon` | `claude-opus-5-5` | Neon, Opus 5.5 |
| Released | 18 November 2026 | 22 September 2026 | Neon, Opus 5.5 |
| Price per 1M tokens (Batch) | $3 input, $15 output ($1.50, $7.50) | $4 input, $20 output ($2, $10) | Neon, Opus 5.5 |
| Context and output | 1,048,576 tokens; 65,536 out per request, about 1M with long decode | 1M tokens; 128K out per request, 300K on the Batch API (beta) | Same pages |
| Default model in | Antigravity | Claude Code | Antigravity changelog, Claude Code docs |
| Also on | Gemini app (Pro, Ultra), Agent Platform, Copilot, Cursor | Amazon Bedrock, Google Cloud, Microsoft Foundry, Claude Platform on AWS | Claude models overview |
| Ahead on (2026-11-20) | Arena WebDev HTML, Vibe Code Bench, Design Arena Overall and 3D Design, cost per task | Arena WebDev overall, Frontend, Gaming and Simulations; Agent Arena | Board table above |
Antigravity's model list also includes Claude Opus 5.5, so one project can use both. Claude Fable 5.2 costs more, $8 and $40 per million tokens (Fable 5.2 docs), and ranks #1 on Arena WebDev at 1,823. See the Fable 5.2 guide and Opus 5.5 vs Sonnet 5.5.
Which Gemini 4 Neon thinking level should you use for UI prototyping and animation?
Medium, by Google's API thinking guide, which lists it for "interactive coding, live UI prototyping", one step below Neon's high default. The Antigravity Neon post of 18 November 2026 instead suggests "quick interactive edits on Low". No Google page recommends a level for 3D or animation code, checked 20 November 2026.
- API and AI Studio: high by default, with low, medium and xhigh available (what's new).
- Antigravity: High by default, Extra (xhigh) above it; Orchestrator's level is fixed (Antigravity models).
- Gemini app: Low, Medium and High only; a hand-picked model starts at Low (Gemini Apps Help).
- Time per level: Artificial Analysis measured 3.7, 8.3, 14.8 and 21.5 minutes per index task from low to xhigh, for scores of 44, 55, 60 and 61 (AA Neon page).
motif's reading, not Google's advice: iterate on layout and timing at medium, run the final one-shot scene at high, and save xhigh for long Three.js builds, where creators report runs of about an hour. To render the code to video, see HTML animation to MP4.
Does motif's archive have Gemini 4 Neon animations yet?
No: none of the 28 Gemini 4 posts in motif's archive on 20 November 2026 names Neon, Carbon or Hydrogen, and its 22 same-prompt Argon runs against Claude are covered in the Argon guide, with archived Opus 5.5 work on the Opus 5.5 page.
Questions
Is Gemini 4 Neon free, and which plan includes it?
No. Neon has no free API tier: every token is billed, at $3 per million input and $15 per million output tokens, checked 20 November 2026. In the Gemini app and Antigravity it needs Google AI Pro ($19.99 a month in the US) or Ultra. Of the four Gemini 4 models, only Hydrogen has a free API tier.
What is the Gemini 4 Neon model ID?
The ID is gemini-4-neon on both the Gemini API and Agent Platform. It is stable only, with no preview or dated alias, as of 20 November 2026. OpenRouter uses the community slug google/gemini-4-neon, and its dated slug ends in 20261118, Neon's 18 November 2026 release date.
What are Gemini 4 Neon's context window and max output?
Neon has a 1,048,576-token context window shared by input, thinking and output. A synchronous request returns at most 65,536 tokens, and long decode reaches 1,048,576 minus the input. Its knowledge cutoff is March 2026. Opus 5.5 caps a request at 128K output tokens. Checked 20 November 2026.
Is Gemini 4 Neon cheaper or better than Claude Opus 5.5?
Cheaper, yes: $3 and $15 per million tokens against $4 and $20 for Opus 5.5, and $6.86 against $12.61 per Artificial Analysis Coding Agent Index task, checked 20 November 2026. Better depends on the job: Opus 5.5 ranks higher on Arena WebDev, 1,818 against 1,763, while Neon scores higher on Vibe Code Bench.
Does Gemini 4 Neon replace Gemini 4 Argon?
In the Gemini app, in practice yes: Google's plan pages swapped Argon for Neon on 18 November 2026. On the Gemini API, Argon is still live with no shutdown date, and Agent Platform calls Neon the target upgrade for Argon workloads. GitHub Copilot is dropping Argon on 14 December 2026.
Can Gemini 4 Neon make animations or videos?
It writes animation code, not video files. Neon outputs text only, so an animation comes out as HTML, Three.js, WebGL2, SVG, CSS or GSAP code that you render to video yourself. Google's 18 November 2026 demos include a WebGL2 ski simulator with about 10,200 lines of shaders.
Can I use Gemini 4 Neon in Gemini CLI, GitHub Copilot or Cursor?
Gemini CLI is not an official Neon channel; its README says it is not tuned for Gemini 4 and points to Antigravity CLI, which added Neon on 18 November 2026 and API-key use on 19 November. GitHub Copilot offers Neon on Pro+, Max, Business and Enterprise. Cursor lists it at $3 input and $15 output on Premium, Super and Ultra.
Is Gemini 4 Neon the new Gemini Pro model?
In upgrade terms, yes: Google lists Neon as the recommended upgrade from Gemini 3.1 Pro Preview, or Carbon for the hardest problems, checked 20 November 2026. There is no Gemini 4 Pro. Google's Gemini 4 models are Argon, Carbon, Hydrogen and Neon, and the Pro name appears only in unofficial posts.
Sources
- Google: Introducing Gemini 4 Neon
- Google DeepMind: Gemini 4 Neon model card
- Google DeepMind: Gemini 4 Neon evaluation methodology (PDF)
- Google AI Studio docs: Gemini 4 Neon model page
- Google AI Studio docs: What's new in Gemini 4 Neon
- Google AI Studio docs: Pricing
- Google AI Studio docs: Changelog
- Google AI Studio docs: Deprecations
- Google AI Studio docs: Choosing a model
- Google AI Studio docs: Thinking
- Google Cloud: Gemini 4 Neon on Agent Platform
- Google Cloud: Model Garden pricing
- Gemini Apps Help: About Gemini models
- Google: Gemini subscriptions (US)
- Google Antigravity: Gemini 4 Neon in Antigravity
- Google Antigravity docs: Changelog
- Google Antigravity docs: Models
- google-gemini/gemini-cli README
- Claude Platform Docs: Claude Opus 5.5
- Claude Platform Docs: Claude Fable 5.2
- Claude Platform Docs: Models overview
- Claude Code Docs: Model configuration
- GitHub Docs: Supported AI models in Copilot
- GitHub Changelog: Gemini 4 Neon in GitHub Copilot
- GitHub Changelog: Upcoming deprecation of selected Copilot models in December
- Cursor Docs: Models and pricing
- OpenRouter: google/gemini-4-neon
- Artificial Analysis: Models leaderboard
- Artificial Analysis: Gemini 4 Neon
- Artificial Analysis: Gemini 4 Neon intelligence gain
- Wayback Machine: Artificial Analysis Gemini 4 Carbon page, 18 November 2026
- Arena: Agent Arena leaderboard
- Arena: Gemini 4 Neon on WebDev
- Arena: WebDev HTML leaderboard
- Arena: WebDev Gaming leaderboard
- Vals AI: Vals Index
- Vals AI: Vibe Code Bench
- Design Arena: Overall leaderboard
- Design Arena: 3D Design leaderboard
- Peter Yang: Gemini 4 Neon vs Claude Opus 5.5 (YouTube)
- Simon Willison: llm-gemini and Gemini 4 Neon
- @thtbee_ on X: Artistic Butterflies in three.js
- @chetaslua on X: Neon motion design in pure JS
- @LuminaBench on X: Neon voxel pagoda
- @kimmonismus on X: early access to Gemini 4 Neon
- @kimmonismus on X: Opus 5.5 remains the best model