Gemini 4 Argon vs Claude Opus 5.5: Is Argon Replaced?

Partly. Checked 20 November 2026, Gemini 4 Argon has no Gemini API shutdown date, but it left the Gemini app on 18 November and GitHub Copilot drops it on 14 December. On 1 December its API price doubles to $4 and $20 per million input and output tokens, matching Claude Opus 5.5, which leads it on WebDev Arena and AA-3D.

AI 辅助写作:本指南由 Claude Sonnet 5.5 起草,每条事实都链接到对应来源。

Is Gemini 4 Argon being replaced?

Not in the API. Google's deprecations page lists gemini-4-argon with "No shutdown date announced" on 20 November 2026, and Agent Platform, formerly Vertex AI, gives a retirement date of "October 13, 2027 or later". Consumer products did drop it: AI Mode on 10 to 11 November and the Gemini app on 18 November 2026.

Google never wrote that another model replaces Argon; its pages stopped listing it. The choosing-a-model page has no Argon row, and the Agent Platform Neon guide calls Neon the "target upgrade" for general coding and enterprise workloads on Argon.

Where can you still pick Gemini 4 Argon?

On 20 November 2026 developers can still pick Argon in seven places: the paid Gemini API, Agent Platform, AI Studio, Antigravity, Android Studio, GitHub Copilot until 14 December 2026, and Cursor. Subscribers can no longer pick it in the Gemini app or AI Mode; the Gemini Apps help page lists only Hydrogen, Neon and Carbon.

Where Gemini 4 Argon can be picked, checked 2026-11-20
SurfaceArgon statusSource
Gemini APIPaid tier only, no shutdown datePricing
Agent Platform (formerly Vertex AI)GA in global, us and eu; retires "October 13, 2027 or later"Argon on Agent Platform
Gemini appRemoved 18 November 2026; Pro and Ultra had it from 13 OctoberHelp page
AI Mode in SearchRemoved 10 to 11 November 2026; Carbon (Ultra only) took its placeAI Mode help
AI StudioIn the playground on AI Pro, Ultra or the paid API tier, per the docsAI Studio plans
Antigravity (app, CLI, IDE extensions)Paid AI Pro, Ultra and Enterprise, plus Orchestrator (Argon); not removedModels, changelog
Android StudioThrough a paid API key or the Antigravity extensionAndroid blog
GitHub CopilotPro+, Max, Business, Enterprise until 14 December 2026, then Carbon. GitHub's decision, not Google'sChangelog, docs
Cursor$2 / $10 per 1M tokens on Premium, Super and UltraCursor pricing
Gemini CLI"Not optimized" for Argon; Google points to AntigravityGemini CLI docs

One Google page lags: deepmind.google still says Argon "is available today for Google AI Pro and Ultra subscribers", contradicting the newer help and plan pages. OpenRouter's google/gemini-4-argon (model list) is a community slug.

How much does Gemini 4 Argon cost after 1 December 2026?

$4 per million input tokens and $20 per million output tokens from 1 December 2026, twice the $2 and $10 Argon costs through 30 November, thinking included, according to Google's pricing page. Cached input goes from $0.10 to $0.20. There is no free tier and no long-context surcharge.

Gemini 4 Argon specs and Gemini API prices, checked 2026-11-20
ItemGemini 4 ArgonSource
Model IDgemini-4-argon, stable only, no preview or aliasModel page
ReleaseAnnounced 30 September 2026 for Fairwind cyber defenders; GA 13 October 2026Announcement, launch post
Context window1,048,576 tokens shared by input and outputModel page
Max output65,536 tokens streaming, 32,768 non-streaming; about 1M minus input with long decodeThinking docs
Thinking levelslow, medium (default), high; no xhighThinking docs
Knowledge cutoffMarch 2026Model card
Standard, per 1M tokens$2 / $10 / $0.10 cached to 30 Nov; $4 / $20 / $0.20 from 1 DecPricing
Batch and Flex$1 / $5, then $2 / $10Pricing
Free tierNone: "not available on the Free tier"What's new in Argon

A prompt of 3,000 input and 12,000 output tokens costs $0.126 now and $0.252 from 1 December 2026. Repricing Artificial Analysis token counts at December list prices, motif gets about $8.63 per Intelligence Index task for Argon at medium against $6.59 for Carbon, because Argon reads more uncached input.

Is Gemini 4 Argon better than Carbon or Neon?

No, on the numbers published so far. Argon and Carbon share one price schedule, and Google says they differ mainly in "how many thinking tokens each model spends per turn" (what's new in Argon). Yet Argon scores 50 on the Artificial Analysis Intelligence Index at high thinking, against 60 for both Carbon and Neon, checked 20 November 2026.

Argon, Carbon and Neon side by side, checked 2026-11-20
ArgonCarbonNeonSource
API price per 1M tokens to 30 Nov$2 / $10$2 / $10$3 / $15Pricing
From 1 Dec 2026$4 / $20$4 / $20$3 / $15Pricing
Default thinking (API)mediummediumhigh, xhigh availableThinking docs
AA Intelligence Index v5.0.150 (high)60 (high)60 (high), 61 (xhigh)Argon, Carbon, Neon
AA output speed68 tok/s51 tok/s110 tok/s (xhigh)Same AA pages
Arena WebDev#18, 1,681#4, 1,769#5, 1,763Arena WebDev

Argon's edges over Carbon are speed, 68 against 51 output tokens per second, and Vals Vibe Code Bench, 91.91% against 89.06%. More on the newer models in Gemini 4 Carbon vs Claude Fable 5.2 and Gemini 4 Neon vs Claude Opus 5.5.

How does Gemini 4 Argon score against Claude Opus 5.5?

It depends on who measures. On Google's Argon model card, with results as of October 2026 and Argon at its highest thinking setting, Argon beats Claude Opus 5.5 on 14 of 18 shared rows. On independent boards checked 20 November 2026, Opus 5.5 is ahead on six of nine, including the Artificial Analysis index, WebDev Arena, Agent Arena and AA-3D.

Google's card has Argon ahead on DeepSWE v1.1 (77.9% vs 74.2%), Vibe Code Bench (91.9% vs 90.3%) and LVBench video understanding (91.7% vs 83.7%), and behind on Terminal-bench 4.0 (57.4% in Google's own run vs 66.4%) and FrontierSWE v2 (55.0% vs 62.3%). Google's later Carbon card uses an Artificial Analysis run for Terminal-Bench 4.0, which gives Opus 5.5 59.6%.

Its methodology PDF takes most rival numbers from public leaderboards such as Vals AI and Terminal-Bench, or from Google's own runs; Opus 5.5's DeepSWE score comes from Anthropic's system card.

Gemini 4 Argon vs Claude Opus 5.5 on independent boards, checked 2026-11-20 (thinking level as each board lists it)
BoardArgon (high)Claude Opus 5.5Leader
AA Intelligence Index v5.0.15059 (max)Opus 5.5
Vals Index v2.2, 18 Nov69.39%, #670.14%, #5Opus 5.5
Vals Vibe Code Bench v1.1, 18 Nov91.91%, #490.29%, #7Argon
Arena Text#2, 1,520#5, 1,506 (high)Argon
Arena WebDev#18, 1,681#2, 1,818 (max)Opus 5.5
Agent Arena, 18 Nov#14, 1,229#2, 1,336 (max)Opus 5.5
AA-3D, Three.js scenes#7, 1147#5, 1202Opus 5.5
Design Arena SVG#5, 1334#8, 1316Argon
Terminal-Bench 4.155.5%65.8% (max)Opus 5.5

Argon's Artificial Analysis score of 53 at launch on 30 September 2026 (AA launch article) was on the older v4 index, so it cannot be compared with today's 50, or 48 at the default medium.

How do Gemini 4 Argon and Claude Opus 5.5 compare on price and specs?

From 1 December 2026 they cost the same per token. Argon's list price becomes $4 input and $20 output per million tokens, exactly the price of Claude Opus 5.5 on the Claude API, and both charge $0.20 for cached reads. Opus 5.5 returns more per request: 128K output tokens against Argon's 65,536.

Gemini 4 Argon and Claude Opus 5.5, checked 2026-11-20
Gemini 4 ArgonClaude Opus 5.5Source
API model IDgemini-4-argonclaude-opus-5-5Argon, Opus 5.5
Price per 1M tokens$2 / $10 to 30 Nov, then $4 / $20$4 / $20Gemini pricing, Claude pricing
Cached input$0.10, then $0.20$0.20 cache readSame pages
Batch$1 / $5, then $2 / $10$2 / $10Same pages
Context window1,048,576 tokens, shared1M tokensModel pages
Max output per request65,536 (32,768 non-streaming)128K; 300K with the Batch betaModel pages
Default reasoningmedium thinkingmedium effort, adaptiveThinking docs, Opus page
Coding tool defaultNo: Antigravity defaults to NeonDefault in Claude CodeAntigravity models, Claude Code docs

Antigravity lists Claude Opus 5.5 and Sonnet 5.5 beside Argon for paid AI Pro, Ultra and Enterprise users (Antigravity models), and has offered them to Agent Platform projects since 15 October 2026 (changelog), so both can run in one tool.

On Claude's side, the Free plan has no Opus 5.5; Pro at $20 a month includes it (Claude pricing).

Is Gemini 4 Argon good for animation, Three.js and SVG?

It depends on the format. On boards checked 20 November 2026, Argon ranks #5 for SVG on Design Arena, above Claude Opus 5.5 at #8. For Three.js scenes and web pages Claude leads: Argon is #7 on AA-3D against Opus 5.5 at #5, and #18 on Arena WebDev against #2.

Design Arena also puts Argon #1 on its agentic image-to-frontend board (Design Arena). Arena WebDev ranks it #16 in Frontend, #15 in Gaming and #20 in HTML, below Fable 5.2, Opus 5.5 and Sonnet 5.5 in Frontend and Gaming (Arena Frontend).

Google's Argon launch post of 13 October 2026 claims better "visual coding" and shows Three.js and GLSL demos such as a desk racer and wizard chess. It never mentions SVG.

Argon outputs text only, so an Argon animation is code you render afterwards; see turn an HTML animation into an MP4.

What do motif's 28 archived Gemini 4 posts show?

As of 20 November 2026, motif's archive holds 28 posts that mention Gemini 4. 23 arrived in launch week: 5 on 13 October, 7 on 14 October and 11 on 15 October 2026. 22 are same-prompt runs of Argon against a Claude model, mostly Claude Opus 5.5, and none names Carbon, Hydrogen or Neon.

motif archive posts that mention Gemini 4, by day posted (motif's own count, archive as of 2026-11-20)
DatePostsSame-prompt runsExample
22 Sep 202610Names "Gemini 4 Pro", which does not exist
5 Oct 202610A 3D story of Claude beating Argon, pre-GA
7 Oct 202610A Japanese explainer about Argon, made with Opus 5.5
13 Oct 202654Opus 5.5 vs Argon in Premiere
14 Oct 202676Remotion product film
15 Oct 20261111Snowy onsen in Blender and Three.js
26 Oct 202611Lobe pump, four models
27 Oct 202610Spookycat, made with Claude Code and Argon
Total282224 creators; 27 posts from X, 1 from Reddit

The Claude side of the 22 runs was Opus 5.5 in 18, Sonnet 5.5 in 2, Haiku 5.5 in 1 and an unnumbered Fable in 1.

What did the posters conclude?

Of the 22 runs, five favour a Claude model on quality, two favour Argon, one creator's four scenes are split, one compares speed only and ten give no verdict in the archived text, which stops at 500 characters.

  • Claude side (5). Le J: "Opus 5.5 is still the king of motion design" (post). A Reddit user: "I lean Opus", though Argon "rebuilt noticeably more of the real UI" (post). Daniel Haida, Julian Goldie and nevian also picked the Claude result.
  • Argon side (2). Web3Bits: "GEMINI 4 ARGON COOKED OPUS 5.5" in a Three.js robot fight (post). Fatal Exit preferred Argon's druid.
  • Split. An Xuan, four scenes: "Argon made the clearer water. Opus made the steam and a night that is actually dark" (post).
  • Speed only. adrolab found Opus 5.5 faster on all four briefs.

What did testers outside the archive find?

  • Hyperautomation Labs (a self-described independent channel that sells Claude Code guides), five prompts on 13 October 2026: Opus 5.5 3, Argon 1, one tie; "Argon for fast, cheap first drafts and visual one-shots" (video).
  • BitsMinds, 15 October: Opus 5.5 12, Argon 8, with Argon run as a Gemini app chat without file tools (build-off).
  • Pat Simmons, 14 October: Argon last and Opus 5.5 first in all four of his builds (AI for Mortals).

Should you still use Gemini 4 Argon?

For new work, Google's own docs now point elsewhere. On 20 November 2026 Argon has no shutdown date in the Gemini API, but Google's choosing-a-model page no longer lists it, its price doubles on 1 December 2026, and at that same price Carbon scores 60 on the Artificial Analysis index against Argon's 50.

Questions

Is Gemini 4 Argon discontinued?

No. On 20 November 2026 Google's deprecations page lists gemini-4-argon with no shutdown date, and Agent Platform gives a retirement date of October 13, 2027 or later. Argon left the Gemini app on 18 November 2026 and AI Mode on 10 to 11 November 2026.

Did Gemini 4 Neon replace Argon?

Only in the Gemini app, and only press says so in words: 9to5Google wrote "Neon also replaces Argon" on 18 November 2026, the day Google's plan lists swapped them. In the API both stay available, and Agent Platform names Neon the target upgrade for Argon workloads.

Is Gemini 4 Argon better than Claude Opus 5.5?

It depends on who measures. Google's model card, with results as of October 2026, has Argon ahead on 14 of 18 shared rows. On independent boards checked 20 November 2026, Opus 5.5 leads on six of nine, including WebDev Arena (1,818 vs 1,681). From 1 December 2026 both cost $4 and $20 per million tokens.

How much does Gemini 4 Argon cost, and is it free?

$2 per million input tokens and $10 per million output tokens through 30 November 2026, then $4 and $20 from 1 December 2026. Cached input goes from $0.10 to $0.20, Batch and Flex cost half, and there is no free tier, checked 20 November 2026.

Which plans include Gemini 4 Argon?

Since 18 November 2026 no Gemini app plan includes Argon. It remains in Antigravity on paid AI Pro, Ultra and Enterprise, in AI Studio on AI Pro, Ultra or the paid API tier, and in GitHub Copilot Pro+, Max, Business and Enterprise until 14 December 2026.

What are Argon's model ID, context window and thinking levels?

The model ID is gemini-4-argon. It has a 1,048,576-token context window shared by input and output, returns up to 65,536 tokens per normal request, offers low, medium and high thinking with medium as default, and has a March 2026 knowledge cutoff, checked 20 November 2026.

Can I use Gemini 4 Argon in Gemini CLI, Cursor or OpenRouter?

Gemini CLI is not optimized for Argon, and Google points to Antigravity instead. Cursor lists Argon at $2 and $10 per million tokens on its Premium, Super and Ultra plans. OpenRouter uses the community slug google/gemini-4-argon. Checked 20 November 2026.

Can Gemini 4 Argon make animations or videos?

It writes animation code. Argon outputs text only, so its animations are HTML, Three.js, WebGL, SVG, CSS or Remotion code rendered to MP4 afterwards, as with Claude Opus 5.5. On 20 November 2026 Argon led Opus 5.5 on Design Arena's SVG board (#5 vs #8), and Opus 5.5 led on AA-3D for Three.js scenes (1202 vs 1147).

来源

  1. Google AI Studio docs: Gemini 4 Argon model page
  2. Google AI Studio docs: Gemini API pricing
  3. Google AI Studio docs: Deprecations
  4. Google AI Studio docs: What's new in Gemini 4 Argon
  5. Google AI Studio docs: Thinking
  6. Google AI Studio docs: Choosing a model
  7. Google AI Studio docs: Google AI plans
  8. Google Cloud: Gemini 4 Argon on Gemini Enterprise Agent Platform
  9. Google Cloud: Gemini 4 Neon guide on Agent Platform
  10. Google: Gemini 4 Argon announcement (30 September 2026)
  11. Google: Introducing Gemini 4 Argon (13 October 2026)
  12. Google: Gemini 4 Carbon now available (10 November 2026)
  13. Google DeepMind: Gemini 4 Argon model card
  14. Google DeepMind: Gemini 4 Argon evaluation methodology (PDF)
  15. Google DeepMind: Gemini models
  16. Gemini Apps Help: About Gemini models
  17. Google Search Help: AI Mode
  18. Wayback Machine: Gemini subscriptions page, 18 November 2026 02:04 UTC
  19. Wayback Machine: Gemini subscriptions page, 18 November 2026 22:18 UTC
  20. Google Antigravity docs: Models
  21. Google Antigravity docs: Changelog
  22. Android Developers Blog: Bringing Gemini 4 Argon to Android
  23. Gemini CLI docs: Model selection
  24. Claude Platform Docs: Claude Opus 5.5
  25. Claude Platform Docs: Pricing
  26. Claude Code Docs: Model configuration
  27. Claude: Plans and pricing
  28. GitHub Changelog: Upcoming deprecation of selected Copilot models in December
  29. GitHub Docs: Supported AI models in Copilot
  30. Cursor Docs: Models and pricing
  31. OpenRouter: Models API
  32. Artificial Analysis: Gemini 4 Argon
  33. Artificial Analysis: Gemini 4 Carbon
  34. Artificial Analysis: Gemini 4 Neon
  35. Artificial Analysis: Claude Opus 5.5
  36. Artificial Analysis: Gemini 4 Argon launch article
  37. Artificial Analysis: AA-3D
  38. Arena: WebDev leaderboard
  39. Arena: Gemini 4 Argon on WebDev
  40. Arena: WebDev Frontend category
  41. Arena: Gemini 4 Argon on Text
  42. Arena: Agent Arena
  43. Vals AI: Vals Index
  44. Vals AI: Vibe Code Bench
  45. Design Arena: SVG leaderboard
  46. Design Arena: Image to Frontend leaderboard
  47. Terminal-Bench: Leaderboard
  48. 9to5Google: Gemini 4 Neon announcement (18 November 2026)
  49. Hyperautomation Labs: Gemini 4 Argon vs Claude Opus 5.5 (YouTube)
  50. BitsMinds: Gemini 4 Argon vs Opus 5.5 vs Astra 6 vs Grok 4.7 build-off
  51. AI for Mortals (Pat Simmons): Gemini 4 Argon review
全部指南