The best AI models right now (November 2026): GPT-6, Claude, Gemini, and Grok compared
Quick answer
As of November 16, 2026, the newest top models are GPT-6 Astra (OpenAI, $10 / $50 per million tokens in / out), Claude Fable 5.2 (Anthropic, $8 / $40), Gemini 4 Carbon (Google, $2 / $10 until November 30), and Grok 4.7 (xAI, $2 / $6). For most coding and writing, Anthropic recommends Claude Opus 5.5 ($4 / $20): Anthropic's Terminal-Bench 4.0 run puts it ahead of GPT-6 Astra, 66.4% to 57.9%, and Google's run puts them level, 59.6% to 59.1%. Each number is one lab's own report, so try 2 or 3 models on your own work.
Free plan: no credit card, 1 published page, 50 form entries a month.
On this page
- The newest flagship from each lab
- GPT-6 vs GPT-5.6: what OpenAI says changed
- GPT-6 vs Claude Opus 5.5
- Claude vs ChatGPT vs Gemini vs Grok for coding
- Claude vs ChatGPT vs Gemini vs Grok for writing
- Free tiers, and the plan that gets the top model
- How to pick a model this month
- Frequently asked questions
- Sources
You're choosing which chat app to pay for this month, or which model your code should call. Since September 3, 2026, OpenAI has released 5 GPT-6 models, Anthropic 4 Claude models, Google 3 Gemini 4 models, and xAI Grok 4.7. This page lines up the newest model from each lab, what it costs in the API and in the app, and what each lab's own benchmarks claim.
Every fact below comes from the lab's own model pages, pricing pages, release notes, or launch posts, read on November 16, 2026 (Japan time). Benchmark numbers are given as the lab that published them reports them, with that lab named. They aren't merged into one leaderboard, because the labs run the same tests in different setups and get different numbers. Where a lab's pages don't say something, the table says so. Prices are in US dollars per million tokens, input then output, unless a row says otherwise.
The newest flagship from each lab
Each lab's most capable model you can use today, checked on November 16, 2026:
| Model (lab) | Released | Context window | API price per 1M tokens (input / output) | App plan with it |
|---|---|---|---|---|
| GPT-6 Astra (OpenAI) | September 3, 2026 | 1,050,000 tokens | $10 / $50 | ChatGPT Pro, from $100 a month; Plus, $20, in Codex |
| Claude Fable 5.2 (Anthropic) | November 9, 2026 | 1M tokens | $8 / $40 | Claude Max, $100, and Ultra, $200; Pro, $20, with usage credits |
| Claude Opus 5.5 (Anthropic) | September 22, 2026 | 1M tokens | $4 / $20 | Claude Pro, $20 a month, and up |
| Gemini 4 Carbon (Google) | November 10, 2026 | 1,048,576 tokens | $2 / $10 until November 30; then $4 / $20 | Google AI Ultra, from $99.99 a month |
| Grok 4.7 (xAI) | September 21, 2026 | 500,000 tokens | $2 / $6 | Premium, $30 a month, lists "Latest Grok models" |
| Muse Wonder (Meta) | Meta doesn't say; in preview | 1,048,576 tokens | $4 / $20 | Every Muse Code plan; no prices listed |
| Mistral Large 4 (Mistral) | November 5, 2026 | 1M tokens | $0.70 / $2.10; open weights | Not named per plan; Pro is $14.99 a month |
What the short cells leave out, from the same pages. GPT-6 Astra takes up to 922,000 input tokens, and a prompt over 272K tokens costs 2x for input and 1.5x for output. Grok 4.7 costs $4 / $12 once a prompt reaches 200K tokens. Claude Fable 5.2 and Opus 5.5 keep one rate across the full 1M. In the ChatGPT app, GPT-6 Astra is Pro's "Pro reasoning" option. Claude Max and Ultra include Fable for up to half the weekly limit, and Claude Pro ($17 a month billed yearly) pays for Fable with usage credits. xAI's plans page doesn't name a model per plan. Muse Wonder Full, Meta's higher-capability variant, comes only with the Power Usage and Maximum Usage plans of Muse Code.
Claude Opus 5.5 is in the table beside Fable 5.2 because Anthropic tells most people to start there: in its November 9 post, Anthropic says, "For most coding and knowledge tasks, we recommend that users start with Claude Opus 5.5, our strongest daily driver." The other labs point the same way. OpenAI's models page says to use GPT-6 Astra "if you're not sure where to start," and GPT-6.1 Sol "to balance intelligence and cost." Google calls Gemini 4 Argon "the ideal choice for the majority of software engineering and knowledge work," and keeps Carbon for the hardest tasks. xAI's models page says, "For everything else, including code, use Grok 4.7."
Each lab also sells a cheaper model one step down:
| Model (lab) | Released | API price per 1M tokens (input / output) | What the lab says it's for |
|---|---|---|---|
| GPT-6.1 Sol (OpenAI) | September 29, 2026 | $2 / $10 | "Near-Astra performance for complex work at a lower cost" |
| GPT-6 Luna (OpenAI) | September 22, 2026 | $0.10 / $0.50 | "Our most efficient model for focused, high-volume tasks" |
| Claude Sonnet 5.5 (Anthropic) | September 28, 2026 | $2 / $10 | "The best combination of speed and intelligence" |
| Claude Haiku 5.5 (Anthropic) | October 7, 2026 | From $0.10 / $0.50 | "High-volume, latency-sensitive tasks such as classification, extraction, and routing" |
| Gemini 4 Argon (Google) | October 13, 2026 | $2 / $10 through November 30; $4 / $20 from December 1 | "Faster responses and lower token spend per turn compared to Gemini 4 Carbon" |
| Gemini 4 Hydrogen (Google) | November 12, 2026 | $1.50 / $7.50, and free on the Gemini API's free tier | "Speed and efficiency for everyday coding and agentic workflows" |
The Chinese labs (DeepSeek, Alibaba's Qwen, Moonshot's Kimi, Zhipu's GLM, and MiniMax) aren't in these tables: their licences, prices, and where you can use them outside China need a comparison of their own: Chinese AI models in November 2026.
GPT-6 vs GPT-5.6: what OpenAI says changed
OpenAI gives each generation several named models. The GPT-5.6 family came out on July 9, 2026: GPT-5.6 Sol "for frontier capability," GPT-5.6 Terra, and GPT-5.6 Luna, with the gpt-5.6 alias pointing to Sol. The GPT-6 family followed: GPT-6 Astra on September 3, GPT-6 Sol and GPT-6 Luna on September 22, GPT-6.1 Sol on September 29, and GPT-6.1 Terra on October 26, per OpenAI's API changelog. So "GPT-6 vs GPT-5.6 Sol" compares GPT-5.6 Sol with either GPT-6 Astra, the top model, or GPT-6.1 Sol, the model in Sol's place.
| GPT-5.6 Sol | GPT-6.1 Sol | GPT-6 Astra | |
|---|---|---|---|
| Released | July 9, 2026 | September 29, 2026 | September 3, 2026 |
| API price per 1M tokens (input / output) | $4 / $20; OpenAI's model page calls this promotional pricing, available until November 30, 2026 | $2 / $10 | $10 / $50 |
| Cached input per 1M tokens | $0.40 | $0.10 | $1 |
| Context window | 1,050,000 tokens | 1,050,000 tokens | 1,050,000 tokens |
| Knowledge cutoff | February 16, 2026 | April 30, 2026 | April 30, 2026 |
Reasoning effort none | Supported | Not supported; use low | Not supported; use low |
| OpenAI's one-line description | "GPT-5.6 flagship model for complex professional work" | "Near-Astra performance for complex work at a lower cost" | "Our most capable model for the most demanding work" |
GPT-6.1 Sol costs half of GPT-5.6 Sol per token, and OpenAI describes it as close to Astra. What OpenAI's GPT-6 guide lists as new for developers:
- Async tool calling. The model keeps reasoning, or answers other parts of a request, while your code runs a tool.
- Mid-turn steering. You send a correction while GPT-6 is working, and the Responses API keeps the work already done.
- Changing reasoning effort mid-conversation without breaking the prompt cache, with a
configuration_updateinput item. - Multi-agent and persisted reasoning on GPT-6.1 Sol. The model hands work to subagents, and can reuse reasoning from earlier turns.
- Misalignment monitoring on GPT-6 Astra. OpenAI's systems watch for misalignment and raise alerts.
The same guide describes how Astra behaves differently. "GPT-6 Astra is generally better than GPT-5.6 Sol and earlier models at staying coherent during long tasks. It is also more likely to ask for clarification where earlier models would make assumptions." OpenAI also says Astra is "stronger at general instruction following," and "can be more sensitive to instructions contained in skills and other files, such as AGENTS.md." If you're moving code over, Astra and GPT-6.1 Sol drop the none reasoning effort, and tool calls on them need the Responses API.
For benchmark numbers on the two generations, the readable source on November 16 was the comparison table in Anthropic's Opus 5.5 post of September 22, which includes both. Anthropic's footnotes say its GPT-6 Astra and GPT-5.6 Sol figures on Terminal-Bench 4.0 are "as reported by OpenAI," and its AutomationBench figures for them come from Zapier's public leaderboard.
| Benchmark (from Anthropic's table, September 22) | GPT-5.6 Sol | GPT-6 Astra |
|---|---|---|
| Terminal-Bench 4.0 (agentic coding) | 37.3% | 57.9% |
| FrontierCode v1.1, main set (agentic coding) | 47.5% | 53.3% |
| Terminal-Bench-Science 0.1 (agentic science) | 22.4% | 64.6% |
| AutomationBench (business workflows) | 28.8% | 41.4% |
| GDPval-AA v2.1 (knowledge work, Elo) | 1588 | 1542 |
In the ChatGPT app, OpenAI's help center says GPT-6 "is based on GPT-6 Sol for Plus, Pro, Business, and Enterprise, and GPT-6 Luna for Free and Go." It adds: "The Pro reasoning option uses GPT-6 Astra."
GPT-6 vs Claude Opus 5.5
The two take prompts and answers of about the same length, and they're far apart in price. Opus 5.5 costs 40% of GPT-6 Astra's price per token, and it keeps that rate across the full 1M context, where Astra's price rises past 272K tokens.
| GPT-6 Astra | Claude Opus 5.5 | |
|---|---|---|
| Released | September 3, 2026 | September 22, 2026 |
| API price per 1M tokens (input / output) | $10 / $50 | $4 / $20 |
| Cache reads per 1M tokens | $1 | $0.20 |
| Context window | 1,050,000 tokens | 1M tokens |
| Maximum output | 128,000 tokens | 128K tokens |
| Long prompts | Over 272K input tokens: 2x input, 1.5x output for the whole request | Same rate across the full 1M |
| Knowledge cutoff | April 30, 2026 | June 2026 (Anthropic's "reliable knowledge cutoff") |
| In the chat app | ChatGPT Pro, from $100 a month | Claude Pro, $20 a month, and above |
The benchmarks depend on who ran them. Anthropic and Google both publish tables that include both models, and on the same test they don't agree:
| Benchmark | Reported by | GPT-6 Astra | Claude Opus 5.5 |
|---|---|---|---|
| Terminal-Bench 4.0 | Anthropic, September 22 (Astra's figure is OpenAI's) | 57.9% | 66.4% |
| Terminal-Bench 4.0, run by Artificial Analysis | Google, Gemini 4 Carbon model card (results as of October 2026) | 59.1% | 59.6% |
| FrontierSWE v2 | Anthropic, November 9; Google's model card gives the same figures | 65.5% | 62.3% |
| FrontierCode 2.0 | Anthropic, November 9 | 42.8% | 45.5% |
| OSWorld 2.1, offline set (computer use) | Anthropic, November 9 | 79.5% | 87.2% |
| OSWorld 2.1, offline subset (computer use) | Google, Gemini 4 Carbon model card | 80.9% | 87.2% |
| Humanity's Last Exam, with tools | Anthropic, September 22 and November 9 | 57.2% | 67.7% |
| Terminal-Bench-Science 0.1 | Anthropic, November 9 | 68.1% | 58.7% |
| Terminal-Bench-Science 0.1, run by Artificial Analysis | Google, Gemini 4 Carbon model card | 63.3% | 59.0% |
| Artificial Analysis Intelligence Index v5.0 | Google, Gemini 4 Carbon model card | 57 | 58 |
Anthropic adds its own caution under its table: "at these levels of capability we've found that benchmark margins have become a less reliable guide to real-world differences." It also reports cost per task. On Terminal-Bench 4.0, Opus 5.5 at its default effort "matches Astra for about 40% of the cost." OpenAI's own GPT-6 Astra launch page returned an error when read on November 16, so OpenAI's benchmark figures appear on this page only where Anthropic quotes them.
If you need more than Opus 5.5, Anthropic's top model, Claude Fable 5.2, scores above GPT-6 Astra on every row of Anthropic's November 9 table that lists both, such as 69.9% to 65.5% on FrontierSWE v2. Fable 5.2 costs $8 / $40, against Astra's $10 / $50.
Claude vs ChatGPT vs Gemini vs Grok for coding
No lab has published a table with all four current flagships. Grok 4.7 came out on September 21, before Opus 5.5, GPT-6.1 Sol, and Gemini 4, so xAI's coding table compares it with GPT-5.6 Sol and Claude Fable 5.1. Here are the coding benchmarks each lab reports, read as they were published:
| Benchmark | Reported by, date | Scores, as reported |
|---|---|---|
| Terminal-Bench 4.0, run by Artificial Analysis | Google, Gemini 4 Carbon model card (results as of October 2026) | Gemini 4 Carbon 65.2%, Claude Opus 5.5 59.6%, GPT-6 Astra 59.1%, Gemini 4 Argon 57.1%, Claude Fable 5.1 52.0% |
| Terminal-Bench 4.0 | Anthropic, September 22 | Claude Opus 5.5 66.4%, GPT-6 Astra 57.9% (OpenAI's figure), Claude Fable 5.1 55.8%, Claude Opus 5 52.3%, GPT-5.6 Sol 37.3% (OpenAI's figure) |
| FrontierSWE v2 | Anthropic, November 9 | Claude Fable 5.2 69.9%, GPT-6 Astra 65.5%, Claude Opus 5.5 62.3%, Gemini 4 Carbon 61.2% |
| FrontierCode 2.0 | Anthropic, November 9 | Claude Fable 5.2 49.1% (xhigh effort), Claude Opus 5.5 45.5%, GPT-6 Astra 42.8%, Gemini 4 Carbon 41.1% (medium effort) |
| CursorBench 4.0 | Anthropic, September 22 | Claude Opus 5.5 57.8%, Claude Fable 5.1 51.8%, Claude Opus 5 46.6%, GPT-5.6 Sol 41.7% |
| CursorBench 4.0 | xAI, September 21 | Claude Fable 5.1 51.8%, Grok 4.7 46.3% (xhigh effort), GPT-5.6 Sol 41.7%, Grok 4.6 40.4% |
| CursorBench 4.0 | Google, November 10 | No score on the page; Google says Carbon "ranks #1 on CursorBench 4.0" at "more than 2X lower cost per task compared to the next best external model" |
What the table supports: on Google's run of Terminal-Bench 4.0, Gemini 4 Carbon leads, and on Anthropic's FrontierSWE v2 and FrontierCode 2.0, Claude Fable 5.2 leads. On these coding rows, Claude Opus 5.5 and GPT-6 Astra are within 4 points of each other, except on Anthropic's own Terminal-Bench 4.0 run, where Opus 5.5 leads by 8.5. On these pages, Grok 4.7 has no coding score beside GPT-6 Astra, Opus 5.5, or Gemini 4.
Price changes the picture for long coding sessions, where the model reads the same files over and over. Cache reads cost $0.20 per million tokens on Claude Opus 5.5 and Claude Fable 5.2, $0.10 on GPT-6.1 Sol, $1 on GPT-6 Astra, and $0.50 on Grok 4.7 under 200K tokens. Anthropic says cache reads "make up the majority of agentic and coding work costs." Each lab also has its own coding agent with its own plans: Codex (OpenAI), Claude Code (Anthropic), Google Antigravity, Grok Build (xAI), and Muse Code (Meta). xAI's docs say Grok 4.7 is also in Cursor, "available on all plans." For how Codex and Claude Code differ in practice, see Codex vs Claude Code skills.
Claude vs ChatGPT vs Gemini vs Grok for writing
None of the launch posts read for this page names a writing benchmark. What each lab says about its model's writing, in its own words:
- Anthropic, Claude Opus 5.5: "Opus 5.5 communicates more naturally than prior models. Early testers found its writing clearer and easier to follow." It "puts the most important information up front." Of Fable 5.2, Anthropic says testers found it communicates "more clearly than past Fable and Mythos models."
- OpenAI, GPT-6 Astra: OpenAI's GPT-6 guide warns that Astra "tends toward detailed, formatted responses and may use recurring phrases across sessions," and "tends to use lists, tables and Markdown." OpenAI describes a cheaper model, GPT-6.1 Terra ($1 / $5), as built for "clear, natural writing."
- Google, Gemini 4 Carbon: Google says Carbon "follows strict schemas and formatting to generate polished and precise spreadsheets, presentations, and reports."
- xAI, Grok 4.7: xAI's docs call it a model "for coding, agentic tasks, and knowledge work," and make no claim about writing.
The nearest measured tests are knowledge-work benchmarks, which grade finished professional work such as reports and analyses. Both tables below are from Anthropic:
| Benchmark (reported by Anthropic) | Scores |
|---|---|
| GDPval-AA v2.1, Elo (September 22) | Claude Opus 5.5 1846, Claude Fable 5.1 1735, Claude Opus 5 1708, GPT-5.6 Sol 1588, GPT-6 Astra 1542 |
| AA-Briefcase v1.1, Elo (November 9) | Claude Opus 5.5 1800, Claude Fable 5.2 1785, Claude Fable 5.1 1669, Gemini 4 Carbon 1611, GPT-6 Astra 1569 |
Writing is the task where your own test matters most. Give 2 or 3 models the same brief, with your own notes and a sample of your style, and compare what comes back.
Free tiers, and the plan that gets the top model
What each chat app gives you for $0, and the plan you need for the lab's top model, from each lab's own pricing pages:
| App | Free plan: the model you get | Next plan up | Plan with the top model |
|---|---|---|---|
| ChatGPT | GPT-6, based on GPT-6 Luna | Go, $8 a month: GPT-6 Luna. Plus, $20: GPT-6 Sol | Pro, from $100 a month: GPT-6 Astra |
| Claude | Sonnet and Haiku | Pro, $20 a month: adds Opus | Max, $100, and Ultra, $200: Fable |
| Gemini | "Our Flash-Lite model" | AI Plus, $4.99 a month: Gemini 4 Hydrogen. AI Pro, $19.99: Gemini 4 Argon | AI Ultra, from $99.99 a month: Gemini 4 Carbon |
| Grok | "Limited" usage; no model named | Plus, $8 a month | Premium, $30: "Latest Grok models" |
| Mistral | "Limited access … including our latest models" | Pro, $14.99 a month | Not named per plan |
The fine print, from the same pages. ChatGPT Free offers "Unlimited text chats with GPT-6," "subject to abuse guardrails," and OpenAI says Go "may include ads." ChatGPT Pro comes at $100, $200, or $500 a month, and GPT-6 Astra is its Pro reasoning option. Claude Pro is $17 a month billed yearly, and pays for Fable with usage credits; Max and Ultra include Fable for up to half the weekly limit. Grok's Ultra, $200 a month, lists "Most powerful Chat models."
For developers, Google's Gemini API has a free tier that includes Gemini 4 Hydrogen at no charge. Gemini 4 Carbon "is not available on the Free tier." The other labs' pages read for this table don't describe a free API tier for their top models.
How to pick a model this month
| If you want | Start with | Why |
|---|---|---|
| One model for most coding and writing | Claude Opus 5.5 ($4 / $20) or GPT-6.1 Sol ($2 / $10) | Anthropic calls Opus 5.5 "our strongest daily driver"; OpenAI calls GPT-6.1 Sol "near-Astra performance for complex work at a lower cost" |
| The hardest, longest jobs | Claude Fable 5.2, GPT-6 Astra, or Gemini 4 Carbon | Each lab's top model. Carbon is the cheapest of the 3: $2 / $10 until November 30, then $4 / $20 |
| The lowest price for high volume | GPT-6 Luna ($0.10 / $0.50) or Claude Haiku 5.5 (from $0.10 / $0.50) | Each lab's model for classification, extraction, and routing |
| A whole codebase in one prompt | A 1M-token model: Claude, GPT-6, Gemini 4, or Muse Wonder | Claude Opus and Fable keep one rate across 1M. GPT-6 Astra charges more past 272K tokens. Grok 4.7 doubles its price past 200K and stops at 500K |
| Weights you can run yourself | Mistral Large 4, Meta's Muse Spark Open, or OpenAI's gpt-oss-120b | Mistral publishes Large 4's weights under its Mistral Ecosystem Open Weight License; Meta calls Muse Spark Open "the most capable US open-weight model to date" |
| A chat app for $0 | ChatGPT Free or Claude Free | ChatGPT Free uses GPT-6 Luna, and Claude Free has Sonnet and Haiku; Gemini Free names only "our Flash-Lite model" |
| A strong model in a chat app for about $20 | Claude Pro, $20 (Opus), or Google AI Pro, $19.99 (Gemini 4 Argon) | ChatGPT Plus, $20, gets GPT-6 Sol in chat; Astra needs Pro |
Before you switch a whole team or a production app, run your own 10 hardest tasks through your top 2 picks at the same effort setting. Anthropic and Google both report cost per task beside their scores, because a model that scores 2 points higher can use far more tokens to get there.
If what you want from the model is a web page, such as a landing page, a sign-up form, or a status page, ask it for one HTML file. Which model writes the best one-page site, on Arena's WebDev boards and in our test: the best AI model for an HTML website. How 9 of these models did with the same landing page prompt, with cost, time, and a phone test: we gave 9 AI models the same landing page prompt. To put that file at a link, see how to run HTML code from ChatGPT, Claude, or Gemini. Convika publishes it at a link that lasts, and a Convika form on the page keeps the entries people leave. Its publish-HTML tool gives a private preview link without an account, and the Free plan publishes 1 page.
Frequently asked questions
What is the best AI model right now?
No single model leads every lab's benchmarks. As of November 16, 2026, the newest top models are GPT-6 Astra (OpenAI), Claude Fable 5.2 (Anthropic), Gemini 4 Carbon (Google), and Grok 4.7 (xAI). On the Artificial Analysis Intelligence Index v5.0, as Google's model card reports it, Gemini 4 Carbon scores 59, Claude Opus 5.5 58, and GPT-6 Astra 57. For everyday coding and writing, Anthropic recommends Claude Opus 5.5, at $4 / $20 per million tokens.
Is GPT-6 better than Claude Opus 5.5?
It depends on whose test you read. Anthropic reports Claude Opus 5.5 at 66.4% on Terminal-Bench 4.0 against GPT-6 Astra's 57.9%, while Google's Artificial Analysis run of the same test puts them at 59.6% and 59.1%. Anthropic's November 9 table puts GPT-6 Astra ahead on FrontierSWE v2, 65.5% to 62.3%. Opus 5.5 costs $4 / $20 per million tokens and GPT-6 Astra $10 / $50, so Opus 5.5 costs 40% as much per token.
What's the difference between GPT-6 Astra, Sol, Terra, and Luna?
They're OpenAI's tiers, from most capable to cheapest. GPT-6 Astra is "our most capable model for the most demanding work," at $10 / $50. GPT-6.1 Sol gives "near-Astra performance for complex work at a lower cost," at $2 / $10. GPT-6.1 Terra is "a fast, efficient model for visual reasoning, precise computer use, and clear, natural writing," at $1 / $5. GPT-6 Luna is "our most efficient model for focused, high-volume tasks," at $0.10 / $0.50.
Is GPT-6 free?
Yes, in a lighter form. ChatGPT's Free plan uses GPT-6 based on GPT-6 Luna, and OpenAI's pricing page lists "Unlimited text chats with GPT-6," subject to abuse guardrails. Plus ($20 a month) uses GPT-6 Sol in chat, and GPT-6 Astra powers the Pro reasoning option on Pro, from $100 a month. In the API, every GPT-6 model is paid per token.
Which AI is best for coding right now?
On Google's Artificial Analysis run of Terminal-Bench 4.0, Gemini 4 Carbon scores highest, at 65.2%. On Anthropic's FrontierSWE v2 and FrontierCode 2.0, Claude Fable 5.2 scores highest, at 69.9% and 49.1%. For a lower price per task, Anthropic recommends Claude Opus 5.5, and OpenAI recommends GPT-6.1 Sol. xAI's Grok 4.7 post compares its coding scores with GPT-5.6 Sol and Claude Fable 5.1, not with these newer models.
Which AI is best for writing?
No lab publishes a writing benchmark. Anthropic says Claude Opus 5.5 "communicates more naturally than prior models," and on the knowledge-work benchmarks Anthropic reports (GDPval-AA v2.1 and AA-Briefcase v1.1), Opus 5.5 has the highest Elo. OpenAI says GPT-6 Astra "tends toward detailed, formatted responses," and built GPT-6.1 Terra for "clear, natural writing." Test 2 or 3 models with the same brief.
Claude vs ChatGPT vs Gemini: which costs least?
For the apps, all three start free. The plan with each lab's top model costs $100 a month on Claude Max (Fable 5.2, for up to half the weekly limit), from $100 on ChatGPT Pro (GPT-6 Astra), and from $99.99 on Google AI Ultra (Gemini 4 Carbon). For about $20 a month, Claude Pro includes Opus 5.5, Google AI Pro ($19.99) includes Gemini 4 Argon, and ChatGPT Plus includes GPT-6 Sol. In the API, Gemini 4 Carbon is the cheapest top model until November 30, at $2 / $10 per million tokens.
Is Gemini 4 better than Gemini 3?
Google's Gemini 4 Carbon model card compares Carbon with Gemini 4 Argon, GPT-6 Astra, and Claude models, not with Gemini 3. Google's models page calls Gemini 3.8 Flash "our previous-generation Flash model," and Google's release notes point Gemini 3.8 Flash users to Gemini 4 Hydrogen, with migration guidance.
Why do the benchmark numbers differ between labs?
Each lab runs the test in its own setup: a different harness, effort level, number of trials, or safeguards. Anthropic's Terminal-Bench 4.0 run gives Claude Opus 5.5 66.4%, and Artificial Analysis's run in Google's model card gives it 59.6%. Anthropic notes its Opus 5.5 scores were run with production safeguards on, which "likely reduces" them on some tests. Compare numbers within one lab's table, not across tables.
Which older AI models are being shut down?
OpenAI, Anthropic, and Google each retire older model ids on a schedule. For every date from November 2026 to February 2027, with the replacement each lab names, see AI model retirements, November 2026 to February 2027.
Sources
Read on November 16, 2026, between 08:29 and 08:45 JST, unless marked. The ChatGPT and Grok pricing pages showed local prices from where they were read, so their US prices come from OpenAI's Codex pricing page and xAI's X Subscription docs.
- OpenAI: models, the GPT-6 Astra, GPT-6.1 Sol, and GPT-5.6 Sol model pages, API pricing, API changelog, Using GPT-6, Codex pricing (ChatGPT plan prices in US dollars), ChatGPT pricing, Intelligent UI in ChatGPT (updated about October 22, 2026). The GPT-6 Astra launch post returned an error.
- Anthropic: models overview, the Claude Fable 5.2, Opus 5.5, Sonnet 5.5, and Haiku 5.5 model pages, API pricing, Claude plans, Introducing Claude Opus 5.5 (September 22, 2026), Introducing Claude Fable 5.2 and Claude Mythos 5.2 (November 9, 2026).
- Google: Gemini API models, Gemini 4 Carbon, What's new in Gemini 4 Carbon, Gemini API pricing, Gemini API release notes, Gemini 4 Carbon model card (published November 2026), Gemini 4 Carbon is now available (November 10, 2026), Google AI plans, US.
- xAI: models and API pricing, Grok 4.7, release notes, The X Subscription prices, Grok plans, Grok 4.7 announcement (September 21, 2026; read through a page summarizer, because x.ai refused direct reads).
- Meta: Muse Wonder, models, pricing and rate limits.
- Mistral: Mistral Large 4, models overview, pricing.
Next step
How to password protect an HTML page or website
A JavaScript password script leaves the password in the page source. To password protect an HTML page or link, encrypt the file or let a server check it.
Related articles
- We gave 9 AI models the same landing page prompt: here's what they builtOne landing page prompt, sent once to GPT-6 Astra, Claude, Gemini 4, Grok 4.7, DeepSeek, Qwen, Kimi and GLM on Nov 16, 2026. Cost, time, phone test.
- Chinese AI models in November 2026: the latest from DeepSeek, Qwen, Kimi, GLM, and MiniMaxEvery major Chinese AI model, from DeepSeek-V4.1 and Qwen4-Max to Kimi K3.1 Pro, GLM-5.3, and MiniMax M3.1: open weights, license, API price, and where to use them.
- What is the Riemann Hypothesis? A plain-English explanationThe Riemann Hypothesis in plain English: what the zeta function and its zeros are, why they control the primes, what is proven, and the November 2026 proof claim.
- Is there a ChatGPT app that publishes a website for you? Yes: Convika, with ChatGPT Sites, Wix, Lovable, and Webflow comparedYes. Convika publishes the website ChatGPT writes, with a form that keeps entries and your own domain: connect https://mcp.convika.com in ChatGPT Business, Enterprise, or Edu, or upload the HTML on any plan.
- ChatGPT, Claude, or Gemini gave me HTML code: how do I run it and see the page?To run HTML code from ChatGPT, Claude, or Gemini, paste the whole code block into a free HTML viewer and it runs in your browser. Or save it as an .html file.
- How to publish or share an HTML page from ChatGPT as a live website, freeHosting for HTML from ChatGPT: compare Convika, Netlify Drop, GitHub Pages, Cloudflare Pages, Tiiny Host, and ChatGPT Sites on free limits, domains, and forms.
Skipbox
Cheap & fast GitHub Actions runners, unlimited for $39 a month
Your own machine keeps Docker layers and package caches between jobs. Switch by changing one runs-on line.
Start a free Skipbox trial