The Model for That

Checked

Which AI should
you use?

Today

Claude Opus 5.5 wins five jobs; Gemini 4 Argon leads chat, research and work

Claude Opus 5.5 wins five of ten jobs and costs $4 in, $20 out per million tokens. Gemini 4 Argon wins three, and its $2/$10 API price doubles after Nov 30. Google announced Gemini 4 Carbon on Oct 28, and only vetted security teams can use it so far.

The model for…

Words, code and thinking

Pictures and video

Use Gemini 4 Argon

For advice, planning and quick answers, Gemini 4 Argon gives the clearest replies right now.

Use Argon in the Gemini app on Google AI Pro ($20/mo). Google keeps it off the Free and AI Plus plans.

Made by Google · $4 per million tokens, blended

The app we’d openGemini gemini.google.com ↗Gemini runs Gemini 4 Argon, our top model for everyday questions, on AI Pro.
Reasoning90Coding91Writing98Vision–Agents77Value42

Runner-up

Claude Opus 5.5

$8 / M tokens

Cheaper

Gemini 3.8 Flash

$1.5 / M tokens

Open weights

MiMo-V2.6-Pro

$0.54 / M tokens · open weights

How the field scores for everyday questionsLMArena text · LMArena instruction following · LMArena multi-turn

Why we picked it Gemini 4 Argon tops the text arena at 1526 and leads multi-turn chat at 1547. Claude Opus 5.5 sits ten points back on text.

  1. 1Gemini 4 ArgonLMArena text 1526 · LMArena instruction following 1528100
  2. 2Claude Opus 5.5LMArena text 1516 · LMArena instruction following 153999
  3. 3Claude Fable 5.1LMArena text 1513 · LMArena instruction following 151997
  4. 4Muse WonderLMArena text 1501 · LMArena instruction following 149397
  5. 5Gemini 3.8 FlashLMArena text 1498 · LMArena instruction following 149295
  6. 6Gemini 3.7 FlashLMArena text 1490 · LMArena instruction following 148692

Get to know the models worth using now

Each petal shows where a model ranks against the rest of this lineup. A long petal means it’s near the top. A dotted petal means nobody has published a score for it yet. The Value petal grows as the price drops.

5 weeks old

Anthropic

Claude Opus 5.5

Claude Opus 5.5 is the best all-round model and wins five jobs, from coding to reading images.

WritingCodingWebsites & apps

Price
$4 in · $20 out
Context
1M
8 weeks old

OpenAI

GPT-6 Astra

GPT-6 Astra is OpenAI's strongest thinker and the pick for hard math and science problems.

Hard problems

Price
$10 in · $50 out
Context
1.1M
4 weeks old

OpenAI

GPT-6.1 Sol

GPT-6.1 Sol matches Astra on hard math for a fifth of the price.

Hard problems

Price
$2 in · $10 out
Context
1.1M
8 weeks old

Anthropic

Claude Fable 5.1

Claude Fable 5.1 reads long documents better than any other model and stays strong at most jobs.

Long documents

Price
$10 in · $50 out
Context
1M
New

Google

Gemini 4 Argon

Gemini 4 Argon is Google's new flagship and gives the best answers for everyday chat and research.

Everyday questionsResearchProfessional work

Price
$2 in · $10 out
Context
1M
4 weeks old

Anthropic

Claude Sonnet 5.5

Claude Sonnet 5.5 does most of what Opus does for half the price, especially agent work.

AgentsWebsites & appsCoding

Price
$2 in · $10 out
Context
1M
New

Meta

Muse Wonder

Muse Wonder is Meta's first top-tier model, and early results show it chats nearly as well as the leaders.

Everyday questionsProfessional work

Price
$4 in · $20 out
Context
1M
8 weeks old

Google

Gemini 3.8 Flash

Gemini 3.8 Flash gives near-flagship answers for a small fraction of the price.

Everyday questionsResearchImages & screenshots

Price
$0.75 in · $3.75 out
Context
1M
Open weights7 weeks old

DeepSeek

DeepSeek V4.1 Flash

DeepSeek V4.1 Flash is the cheapest strong open model and builds websites well.

Websites & appsCoding

Price
$0.045 in · $0.50 out
Context
1M
NewOpen weights

Meta

Muse Spark Open

Muse Spark Open is Meta's new open-weight coding model, and early results put it mid-pack.

Research

Price
$1.25 in · $4.25 out
Context
1M
Open weights5 weeks old

Xiaomi

MiMo-V2.6-Pro

MiMo-V2.6-Pro is Xiaomi's open model and the strongest open pick for most jobs.

Everyday questionsWritingCoding

Price
$0.43 in · $0.87 out
Context
1M
New

OpenAI

GPT-6.1 Terra

GPT-6.1 Terra is OpenAI's new mid-tier model for computer use, and its early math scores are strong.

Hard problems

Price
$1 in · $5 out
Context
1.1M
Announced

Google

Gemini 4 Carbon

Gemini 4 Carbon isn't public yet, and Google reports it tops one independent intelligence index.

Price
–
Context
–
New

Moonshot

Kimi K3.1 Pro

Kimi K3.1 Pro is Moonshot's newest model, and early tests put it near the top for coding and sites.

Websites & appsCoding

Price
$2 in · $10 out
Context
1M

Meet the apps that put these models to work

Most people reach these models through an app. Some apps, like Perplexity, run models from several labs and add their own search on top. Leaderboards don’t rank apps, so Claude picks these from its own web research each time it rewrites the guide.

Search

Perplexity

Perplexity

Perplexity searches the web and answers with sources you can click and check.

ResearchEveryday questions

Search

Google

Google AI Mode

Google AI Mode answers your searches inside Google, with Gemini 4 Argon for paid subscribers.

ResearchEveryday questions

Costs
AI Pro $19.99/mo for Argon
Runs on
Gemini 4 Argon on AI Pro and Ultra
Assistant

OpenAI

ChatGPT

ChatGPT is the all-round assistant with the most features, and Plus runs GPT-6 Sol.

Everyday questionsHard problemsMaking images

Costs
Free; Go $8, Plus $20, Pro from $100/mo
Runs on
GPT-6 Sol on Plus, GPT-6 Luna on Free, GPT-6.1 Terra in Work and Codex
Assistant

Anthropic

Claude

Claude is Anthropic's app for writing and long documents, with Opus 5.5 on paid plans.

WritingLong documentsImages & screenshots

Costs
Free; Pro $20/mo, Max $100 or $200/mo
Runs on
Claude Opus 5.5 on paid plans, plus Anthropic's other Claude models
Assistant

Google

Gemini

Gemini is the home of Gemini 4 Argon, our pick for everyday questions and work.

Everyday questionsProfessional workResearch

Costs
Free; AI Plus $4.99, AI Pro $19.99/mo
Runs on
Gemini 4 Argon, Gemini 3.8 Flash, Nano Banana 2.1
Assistant

xAI

Grok

Grok is a handy second app for live news and posts from X.

Everyday questions

Costs
Free; X Subscription Plus $8, Premium $30/mo
Runs on
Grok 4.3 by default
Assistant

Microsoft

Microsoft Copilot

Microsoft Copilot is the AI for people who work in Microsoft 365.

Professional work

Costs
Free; Microsoft 365 Premium $19.99/mo
Assistant

Meta

Meta AI

Meta AI answers questions inside WhatsApp, Instagram and Messenger.

Everyday questions

Costs
Free
Runs on
Muse Spark 1.3
Coding

Anthropic

Claude Code

Claude Code puts Opus 5.5 to work inside your codebase and terminal.

CodingAgentsWebsites & apps

Costs
No free tier; Pro $20/mo, Max $100 or $200/mo
Runs on
Claude Opus 5.5, Sonnet 5.5, Haiku 5.5
Coding

OpenAI

Codex

Codex is OpenAI's coding agent, and it works through tasks in the cloud on its own.

CodingAgents

Costs
Free with ChatGPT; Plus $20/mo, Pro from $100
Runs on
GPT-6.1 Terra, GPT-6 Astra, GPT-6 Sol, GPT-6 Luna
Coding

Anysphere

Cursor

Cursor is an AI code editor for day-to-day work in your own project.

Coding

Costs
Free; Plus $8, Premium $30/mo
Runs on
Cursor's own models and Grok; Claude, GPT, Gemini at API price
Coding

GitHub

GitHub Copilot

GitHub Copilot works inside the GitHub tools and editors you already use.

Coding

Costs
Free; Pro $10/mo, Pro+ $39/mo
Runs on
Your pick of models, incl. GPT-6.1 Terra and Gemini 4 Argon
Site builder

Lovable

Lovable

Lovable turns a plain description into a working web app.

Websites & apps

Costs
Free 5 credits a day; Pro from $25/mo
Site builder

Vercel

v0

v0 is Vercel's builder for turning a prompt into a working site.

Websites & apps

Costs
Free $5 credits/mo; Team $30 per user
Site builder

StackBlitz

Bolt

Bolt builds a working web app from a prompt and bills you by tokens.

Websites & apps

Costs
Free 1M tokens/mo; Pro from $25/mo
Site builder

Replit

Replit

Replit's Agent builds an app for you and keeps working with little hand-holding.

Websites & appsAgents

Costs
Free with limited AI; Core $20/mo
Agent

xAI

Grok Bot

Grok Bot gives you AI teammates that keep working on a cloud computer after you log off.

Agents

Costs
In SuperGrok Heavy, Cursor Ultra or Teams Premium
Images

Midjourney

Midjourney

Midjourney makes artistic, cinematic images and concept art.

Making images

Costs
Basic $10/mo; Standard $30/mo
Runs on
Midjourney V8.2 by default, older versions selectable
Images

Ideogram

Ideogram

Ideogram makes posters, logos and thumbnails with clean text in the image.

Making images

Runs on
Ideogram 4.5 and open-weight Ideogram 4.0
Images

Krea

Krea

Krea puts many image and video models in one place with fast, real-time editing.

Editing photosMaking images

Runs on
Krea 3, Nano Banana Pro, Flux, Ideogram, Veo, Kling and more
Images

Adobe

Photoshop

Photoshop is the desktop tool for precise retouching with AI fills built in.

Editing photos

Costs
7-day trial; Photoshop $22.99/mo
Runs on
Firefly Image 4, plus partner models like Nano Banana
Video

Runway

Runway

Runway is a video studio for creative control, editing clips and making ads.

Making videoEditing photos

Costs
Free credits; Standard from $12/mo
Runs on
Gen-4.5, Aleph 2.0, Seedance 2.0, Seedream 5.0
Video

Kuaishou

Kling

Kling makes realistic video of people in motion at a good price.

Making video

Costs
Free 66 credits a day; Standard $10/mo
Runs on
Kling 4.0, 3.0 Omni, 3.0 Turbo
Video

Google

Flow

Flow is Google's video studio, and it runs Veo 3.1 to turn prompts into clips.

Making video

Costs
Free credits; AI Pro $19.99/mo gets 1,000
Runs on
Veo 3.1, Gemini Omni Flash

Meet the companies behind them

7 models

Led by Sam Altman

OpenAI

OpenAI makes ChatGPT and Codex. It moved ChatGPT to GPT-6 in October, and GPT-6.1 Terra now handles computer tasks for paid users at half the old price.

Flagship model
GPT-6 Astra
Apps
ChatGPT, Codex
4 models

Led by Dario Amodei

Anthropic

Anthropic makes Claude and Claude Code. Opus 5.5 arrived in September and still tops several independent rankings, with Sonnet 5.5 and Haiku 5.5 following.

Flagship model
Claude Opus 5.5
Apps
Claude, Claude Code
5 models

Led by Sundar Pichai

Google

Google makes Gemini, NotebookLM and AI Mode in Search. Gemini 4 Argon reached Pro subscribers in October, and Gemini 4 Carbon is going only to security teams for now.

Flagship model
Gemini 4 Argon
Apps
Google AI Mode, Gemini, Flow
1 model

Led by Elon Musk

xAI

xAI makes the Grok chatbot, which reads live posts from X. It now calls itself SpaceXAI and faces a reopened police inquiry over Grok deepfakes.

Flagship model
Grok 4.6
Apps
Grok, Grok Bot
3 models

Led by Mark Zuckerberg

Meta

Meta makes the Muse models and the Meta AI app. Muse Wonder launched in October for paid Muse plans, and Muse Spark Open is free to download.

Flagship model
Muse Wonder
Apps
Meta AI

Led by Satya Nadella

Microsoft

Microsoft sells Copilot for Windows and Office, plus GitHub Copilot for coders. GitHub Copilot lets you pick models from other labs and added GPT-6.1 Terra in October.

Apps
GitHub Copilot, Microsoft Copilot
3 models

Led by Liang Wenfeng

DeepSeek

DeepSeek makes cheap open models and a chat app. Its October update to V4.1 Flash codes well for little money, and older V4 Flash was retired.

Flagship model
DeepSeek V4.1 Flash
1 model

Led by Eddie Wu

Alibaba

Alibaba makes the Qwen family of models, and Qwen3.8 Max is its current top model.

Flagship model
Qwen3.8 Max
3 models

Led by Yang Zhilin

Moonshot AI

Moonshot AI makes the Kimi models and apps. Kimi K3.1 Pro came out in October and sits close to top US models, with open weights promised soon.

Flagship model
Kimi K3
1 model

Led by Arthur Mensch

Mistral AI

Mistral AI is a Paris lab that makes the Le Chat assistant.

Flagship model
Mistral Large 4

Led by Aravind Srinivas

Perplexity

Perplexity runs an answer engine that searches the web and cites sources. It uses models from other labs and has no model of its own on our list.

Apps
Perplexity
1 model

Led by Jensen Huang

NVIDIA

NVIDIA makes the chips most AI runs on and also releases open models. Nemotron 3.5 Super VL reads images and video and leads US open models of its size.

Flagship model
Nemotron 3.5 Super VL
2 models

Led by Zhang Peng

Z.ai

Z.ai is a Beijing lab that makes the GLM models. GLM-5.3 is its current model, and a smaller Flash version costs less.

Flagship model
GLM-5.3
1 model

Led by Yan Junjie

MiniMax

MiniMax is a Shanghai lab with cheap open models. MiniMax-M3.1-Flash costs very little, but reviewers say it makes things up often.

Flagship model
MiniMax-M3.1-Flash

See what each model costs for what it can do

Higher means smarter and further left means cheaper. The line traces the best model you can buy at each price. Anything below it costs more than it should.

Measure smarts by
Smart and cheap$0.1$1$10$100120125130135140145150155160165170Price per million tokens (blend of input and output, log scale) →Epoch Capabilities Index →Claude Opus 5.5GPT-6 AstraGPT-6.1 SolClaude Fable 5.1Gemini 4 ArgonClaude Sonnet 5.5Muse WonderClaude Haiku 5.5Gemini 3.8 FlashDeepSeek V4.1 FlashMuse Spark OpenQwen3.7 Flashgpt-oss-20b

In the lineup Other current models Open weights Best you can get at each price

Catch up on what changed this month

  1. Google announces Gemini 4 Carbon for vetted security teams

    Selected cyber defenders get Carbon this week, and the public gets it later with no date. Google says it ranks first on the Artificial Analysis index, and Vals puts it second overall.

    Gemini 4 Carbon

  2. OpenAI launches GPT-6.1 Terra at half the old Terra price

    GPT-6.1 Terra costs $1 in, $5 out and can operate your computer for you. It runs in ChatGPT Plus and higher, Codex and GitHub Copilot, and Terra Pro is the same model in a pro reasoning mode.

    GPT-6.1 TerraGPT-5.6 TerraGPT-6.1 Terra Pro

  3. Moonshot releases Kimi K3.1 Pro and a faster Highspeed version

    K3.1 Pro costs about a third less than K3 and scores near Opus 5.5 on terminal coding. Moonshot promises the weights by Nov 4.

    Kimi K3.1 ProKimi K3

  4. MiniMax ships M3.1 Flash with open weights

    The model costs $0.15 in, $0.60 out, and its free OpenRouter period ended Oct 22. VentureBeat flagged a high hallucination rate.

    MiniMax-M3.1-Flash

  5. DeepSeek updates V4.1 Flash with faster, shorter answers

    V4.1-Flash-1017 ships open weights under MIT and keeps its prices. DeepSeek retired the older V4 Flash on Sep 10.

    DeepSeek V4.1 FlashDeepSeek V4 Flash

  6. Meta and NVIDIA release new open-weight models

    Meta's Muse Spark Open targets coding and runs at home on two Mac Studios. NVIDIA's Nemotron 3.5 Super VL reads images, documents and video at very low prices.

    Muse Spark OpenNemotron 3.5 Super VL

  7. Gemini 4 Argon opens to developers and AI Pro subscribers

    Google announced Argon on Sep 30 for Ultra and added Pro on Oct 13. Its API price of $2 in, $10 out doubles after Nov 30.

    Gemini 4 Argon

  8. Meta launches Muse Wonder, its first top-tier model

    Muse Wonder remembers you across chats and needs a paid Muse plan. A bigger Wonder Full runs only in Muse Code on the $50 and $200 plans.

    Muse Wonder

Check the raw rankings yourself

  1. 1Gemini 4 Argon1526
  2. 2Claude Opus 5.51516
  3. 3Claude Fable 5.11513
  4. 4Claude Opus 51508
  5. 6Claude Opus 4.61501
  6. 7Muse Wonder1501
  7. 8Gemini 3.8 Flash1498
  8. 10Muse Spark 1.31491
  9. 11MiMo-V2.6-Pro1491
  10. 12Claude Fable 51491
  11. 13Claude Opus 4.71490
  12. 14Gemini 3.7 Flash1490

From LMArena, where people compare two anonymous answers and vote for the better one. Ratings work like chess Elo: 100 points apart means the higher model wins about 64% of matchups. Board published 2026-10-22.