Skip to main content
BenchLM
Data

The best AI chatbot, scored on what you can actually buy

Our leaderboard says Claude wins by a mile. Limited to the models each $20 plan includes, Claude Pro leads ChatGPT Plus by 0.9 points, and Claude's free plan beats Google's $19.99 one.

Two hand-inked bars on dusk-blue ledger paper, running right from a pencil line: a long black bar above, and below it a short cobalt bar ending in a cobalt stain, about a fifth as long
In this article5 sections

For most people, the best AI chatbot as of November 13, 2026 is Claude, but by far less than our leaderboard suggests.

Ranked by the best model each company sells, Claude wins easily. Ranked by what a $20 plan gets you, Claude Pro scores 86.6, ChatGPT Plus 85.7, and Google AI Pro 79.0.

We found that out by breaking our own comparison.

Our LLM selector scores ChatGPT, Claude, and Gemini on 15 groups of jobs, and it showed Claude ahead in 8 of them. Then we checked which models those scores came from. Most of Claude's came from a model that Claude Pro doesn't include.

So we reran the whole table with one rule: each app may use only the models its plan lists. The gap at the top shrank from 4.5 points to 0.9. Claude's free plan finished ahead of Google's $19.99 one. And the useful question changed from "which chatbot is best?" to "which plan includes a model that's good at my job?"

The leaderboard ranks models a $20 plan doesn't include

Here's what each company's plans include, scored on our overall score. The overall score combines the benchmarks we track into one number per model, on the November 13 data.

Overall scores on November 13, 2026 data. Claude: free plan 83.1 (Sonnet 5.5), Pro at $20 86.6 (Opus 5.5), best model 93.7 (Fable 5.2, on Max at $100). ChatGPT: Plus at $20 85.7 (GPT-6 Astra), also its best model. Gemini: Google AI Pro at $19.99 79.0 (Gemini 4 Argon), best model 89.2 (Gemini 4 Carbon, on Ultra from $99.99).

Claude's best model is Claude Fable 5.2, at 93.7. Claude Pro lists Opus, Sonnet, and Haiku in its allowance and puts Fable under usage credits, billed at API rates on top of the $20. Fable is included on Max, at $100 a month, for up to half the weekly limit. Gemini's best, Gemini 4 Carbon at 89.2, is listed only on Google AI Ultra, from $99.99. ChatGPT is the exception. Its best model, GPT-6 Astra at 85.7, is in the $20 Plus plan.

You might expect cost to explain who gets rationed. It doesn't. Through the API, GPT-6 Astra costs $10 per million input tokens and $50 per million output tokens. That's more than Fable 5.2 ($8 and $40) and five times Gemini 4 Carbon ($2 and $10). The most expensive model to run is the one in a $20 plan.

What lines up is age. Fable 5.2 came out on November 9, 2026, and Carbon became generally available on Google's API on November 10. Astra came out on September 3. We can't see the companies' reasons, but the pattern in their plan pages is consistent: a new flagship starts on the top-priced plan, and the $20 plan gets an older or smaller model.

That's worth knowing whenever you read any AI leaderboard, ours included. The top row is usually a model that's days or weeks old, and most people can't use it without paying about five times more. In our data the gap between what a company sells and what $20 buys was 7.1 points for Claude, 10.2 for Gemini, and zero for ChatGPT.

At $20 a month, each plan wins different jobs

With each app limited to its plan's models, here's how the 15 job groups split. A single benchmark counts as a tie when the results are within 1.5 points. The full job-by-job table, with the model behind each score, is on the LLM selector.

Table 1
Plan, about $20 a month Wins outright Ties for the lead
Claude Pro (Opus 5.5, Sonnet 5.5) Coding (82.6 vs 78.7), agents and slides (83.8 vs 74.9), studying (89.8 vs 85.6), data analysis (57.5 vs 51.2), research (48.1 vs 46.1), overall (86.6 vs 85.7) Images, PDFs, Excel, finance, data extraction, website building
ChatGPT Plus (GPT-6 Astra) Math (93.7% vs 91.2%), reasoning (81.1 vs 80.5) Images, PDFs, data extraction
Google AI Pro (Gemini 4 Argon) Legal research (76.8% vs 72.4%) Excel, finance, website building

The first number in each pair is the winner's score. The second is the runner-up's.

One of Claude's wins and one of ChatGPT's are close enough that we'd call them draws. Claude leads overall by 0.9 points and ChatGPT leads reasoning by 0.6. Those two rows are category scores, where we don't apply a tie band, but a gap under a point shouldn't decide a subscription.

One caveat changes ChatGPT's column. OpenAI's plan page lists some Plus models for Codex or ChatGPT Work, not the chat window. GPT-6.1 Sol, which gives ChatGPT its tie in data extraction, is listed for Work and Codex, not Chat. If you only ever type into the chat box, ChatGPT Plus is a little weaker than its row suggests.

Gemini drops furthest because Google's $20 plan stops at Gemini 4 Argon, which scores below Carbon on 8 of the 15 rows. Google AI Pro also lists a Pro model and a Flash model without version numbers, so we can't score them.

The best free AI chatbot beats some $20 plans

Claude Free runs Sonnet and Haiku, and Claude Sonnet 5.5 has an overall score of 83.1. That's 4.1 points above Google AI Pro at $19.99. Against ChatGPT Plus at $20, Sonnet 5.5 scores higher on coding (81.0 vs 78.7), website building (92.4% vs 89.6%), and data analysis (57.5 vs 51.2).

The free plan is limited by how much you can use it, not by how good the model is. Anthropic's plan page lists limited usage every five hours and every week on Free, and doesn't publish the number. What you pay $20 for is mostly more turns with a model that's 3.5 points better.

ChatGPT Free and Gemini Free are harder to judge, because neither plan page names the version of the chat model. OpenAI lists unlimited text chats with GPT-6 on Free without saying which GPT-6. The model it does name on Free, for the Codex coding agent, is GPT-6 Luna, which scores 61.1 overall. We don't assume the chat window runs Luna. Gemini Free lists "our Flash-Lite model", again without a version.

What the free plans include differs more than their models do. Gemini Free lists deep research, image generation and editing, Gemini Live voice chat, and 15 GB of storage. ChatGPT Free lists deep research, image creation, and voice, with limits. Claude Free has voice and memory but no deep research, which starts on Pro.

So, if you aren't paying: Claude Free has the strongest model we can verify, and Gemini Free has the most tools.

Three questions to ask before paying for any AI assistant

Everything above comes down to three checks. They work for any chatbot, including ones we don't score, and they'll still apply after the next model release.

1. Which model does this plan actually include? Read the plan page, not the launch post. A plan that lists a family name without a version, as ChatGPT Free and Gemini Free do, doesn't tell you what you're getting. Microsoft Copilot goes further: Microsoft says the app uses "models from Microsoft, OpenAI, and other model developers" and names none, so no benchmark can be attached to it. Its case is placement instead. Microsoft 365 Personal, at $9.99 a month, puts Copilot in Word, Excel, PowerPoint, Outlook, and OneNote.

2. Which job will you use it for most? The table above shows the answer depends on it. If you mostly code, write, or study, Claude Pro is our pick. For math, ChatGPT Plus is ahead by 2.5 points. For legal research, Google AI Pro leads by 4.4. Coding has a longer answer, with cost per fixed bug, in ChatGPT or Claude for coding.

3. What happens to your chats? This differs more than the scores do. On personal plans, as of the pages we read on November 10, 2026:

Table 2
ChatGPT Claude Gemini Copilot
Chats train the models Yes, by default Only if you allow it, or if a chat is flagged for safety review Yes while Keep Activity is on, and reviewers read a subset No, per Microsoft
How to stop it Turn off "Improve the model for everyone" Turn off model improvement Turn off Keep Activity Nothing to turn off
Ads The $8 Go plan "may include ads" Not mentioned Google says chats aren't used for ads Possible without Microsoft 365

If privacy matters more to you than a point of score, weigh this table above the first one.

Two apps aren't in our tables. xAI's newest model, Grok 4.7, has no scored results on our data as of November 13, so we left Grok out; the model rows we have are on Grok vs Claude. Perplexity runs other companies' models, and we don't track which one its app uses. More options are on our ChatGPT alternatives page.

Run a ten-minute test before you pay

Every score here comes from a fixed benchmark, run by its publisher or an outside lab, not from your work. Two plans within a point of each other can swap places on your questions. Here's a test that costs nothing.

  1. Pick three real tasks from last week: one you'd write, one you'd research, and one with a file or a table.
  2. Give the same three prompts to the free tiers of two apps. Claude Free and ChatGPT Free are the obvious pair.
  3. For each answer, check one fact you already know, note what it left out, and count how many follow-ups it took to get something usable.

If one app needs fewer follow-ups on two of the three tasks, that's your answer, whatever the leaderboard says. If they tie, choose on the privacy table and the tools you'll actually use.

The research score is worth a look while you test. The AA-Omniscience Index, from Artificial Analysis, gives points for right answers to knowledge questions, takes points away for wrong ones, and treats "I don't know" as neutral. Artificial Analysis says scores run from -100 to 100, and "negative scores mean more incorrect than correct." At $20, Opus 5.5 scores 48.1 and GPT-6 Astra 42.4. GPT-6 Luna, the model OpenAI names for Codex on ChatGPT Free, scores -8.3. A model that guesses when it doesn't know is the one to double-check.

We'll rerun the plan comparison when one of three things happens: Anthropic moves Fable into the Pro allowance, Google brings Carbon below $99.99, or Grok 4.7 gets its first scored results. The full model ranking is on best overall, and the app-by-app pages are ChatGPT vs Claude, ChatGPT vs Gemini, and Claude vs Gemini.

Frequently asked questions

01Is ChatGPT or Claude better for everyday questions?

On the $20 plans, it's close to a draw. Claude Opus 5.5, on Claude Pro, has an overall score of 86.6 on November 13, 2026 data, and GPT-6 Astra, on ChatGPT Plus, has 85.7. A gap under a point shouldn't decide a subscription, so let features, privacy settings, and a test on your own work settle it.

02Which AI chatbot is best for free?

Claude Free, on the models we can score. It runs Claude Sonnet 5.5, with an overall score of 83.1 on November 13, 2026 data. ChatGPT Free and Gemini Free don't name the model version their chat uses. Gemini Free lists more tools, including deep research, which Claude Free leaves out.

03Which AI chatbot makes things up least?

On the AA-Omniscience Index, a knowledge test from Artificial Analysis that takes points away for wrong answers, Claude Opus 5.5 scores 48.1, Gemini 4 Argon 46.1, and GPT-6 Astra 42.4. Below zero means more wrong answers than right ones. GPT-6 Luna, the model on ChatGPT Free's coding agent, scores -8.3.

04Is Microsoft Copilot as good as ChatGPT?

We can't say from benchmarks. Microsoft doesn't name the model the Copilot app runs, only that it uses models from Microsoft, OpenAI, and others, so no score can be attached to it. Copilot's case is the Office apps: Microsoft 365 Personal, at $9.99 a month, puts it inside Word, Excel, PowerPoint, Outlook, and OneNote.

Reader questions

Share or save

Share on XShare on LinkedIn

Keep readingAll research

Choose the right model before an expensive mistake. One weekly recommendation: what to choose, what costs less, and what is not worth switching to.

Join 7,000+ subscribers.