Data-ranked, updated every 2 days
Best AI Chatbot (setembro de 2026)
ChatGPT is the best AI chatbot right now, ranking first of 7 on our composite score with 94 out of 100 (setembro de 2026 usage data, LMArena board of 8 de outubro de 2026). It drives 80,5% of AI chatbot referral traffic worldwide (rank 1). Its best model, gpt-5.6-sol-xhigh, has an LMArena blind-vote rating of 1485. Gemini is second with 79, 15 points behind.
Usage: StatCounter referral share, setembro de 2026. Quality: LMArena blind votes, board of 8 de outubro de 2026. Page refreshed 13 de outubro de 2026.
The ranking: 7 AI chatbots, scored
One row per consumer assistant. Each rating is the best model of that vendor on the LMArena board, with the model name beside it. Ranked on the composite score out of 100; the method is below the table.
| # | Chatbot | Score | Usage share | Quality (LMArena text) | Web search (LMArena) | Coding (LMArena WebDev) | Citations per answer |
|---|---|---|---|---|---|---|---|
| 1 | ChatGPT | 94 | 80,5% | 1485 gpt-5.6-sol-xhigh | 1256 gpt-5.6-sol-xhigh | 1786 gpt-6-astra-max | 6.8 |
| 2 | Gemini | 79 | 10,9% | 1525 gemini-4-argon-high | 1208 gemini-3.1-pro-grounding | 1678 gemini-4-argon-high | 1.9 |
| 3 | Claude | 71 | 1,0% | 1507 claude-opus-5.5-high | 1229 claude-fable-5 | 1813 claude-opus-5.5-max | 6.6 |
| 4 | Perplexity | 65 | 5,8% | n/a | 1143 ppl-sonar-pro-high | n/a | 22.4 |
| 5 | Meta AI | 56 | n/a | 1494 muse-spark-1.3-max | n/a | 1657 muse-spark-1.3-max | n/a |
| 6 | Grok | 55 | n/a | 1475 grok-4.20-beta1 | 1205 grok-4.20-multi-agent-beta-0309 | 1639 grok-4.7-xhigh | 56.8 |
| 7 | DeepSeek | 52 | 0,0% | 1475 deepseek-v4.1-flash-max | n/a | 1619 deepseek-v4.1-flash-max | 3.0 |
| n/a | Microsoft Copilot | not ranked | 1,8% | n/a | n/a | n/a | n/a |
Usage is the share of website referrals worldwide, not a user count. A chatbot needs at least 2 of the 4 scored signals to be ranked; others are listed below the ranking with the data they have. n/a means the source does not list that chatbot.
- Perplexity: no data for quality, coding, counted at the lowest score any listed chatbot has.
- Meta AI: no data for usage, web search, counted at the lowest score any listed chatbot has.
- Grok: no data for usage, counted at the lowest score any listed chatbot has.
- DeepSeek: no data for web search, counted at the lowest score any listed chatbot has.
Citations per answer: average web sources cited per answer in the Essentras AI Answer Index, 1922 answers across 105 checks in the last 90 days. Shown for context, not part of the score.
How we ranked them
Every number on this page comes from three data sources: StatCounter Global Stats (AI chatbot referral share, latest complete month: setembro de 2026), LMArena blind-vote leaderboards (text board of 8 de outubro de 2026, search board of 24 de agosto de 2026) and the Essentras AI Answer Index. Nothing is estimated or invented.
- Quality, weight 40%: the highest LMArena text-arena rating among that vendor's models.
- Usage, weight 30%: StatCounter worldwide referral share for the latest complete month.
- Web search, weight 15%: the highest rating among that vendor's models in the LMArena search arena.
- Coding, weight 15%: the highest rating among that vendor's models in the LMArena WebDev arena.
Ratings are turned into a score out of 100 with the same Elo math LMArena uses: a model's chance of beating the top-rated model in a blind vote, doubled, so the leader scores 100 and a model that wins 43% of head-to-heads scores 86. Usage share runs from under 0.1% to over 80%, so it is scored on a log scale against the leader. The score is the weighted sum.
When a chatbot has no value for a signal (for example, StatCounter does not track it), that signal counts at the lowest score any listed chatbot has for it, so missing data never raises a rank. Weights are the same for every chatbot. A chatbot with fewer than 2 signals is not ranked.
LMArena lists models by organization: OpenAI for ChatGPT, Google for Gemini, Anthropic for Claude, xAI for Grok, DeepSeek for DeepSeek, Meta for Meta AI and Perplexity for Perplexity. Microsoft Copilot has no quality, search or coding rating because we could not verify at a Microsoft page which models it runs, and we do not borrow another vendor's score.
Limits: arena ratings measure what voters prefer in blind comparisons, usage measures referral traffic and not users, and a chatbot app can run a different model from the one tested. Treat close scores as ties.
Best AI chatbot for each job
Each pick is computed from the same data as the ranking.
- Best overallChatGPTChatGPT has the highest composite score of the 7 ranked chatbots, 94 out of 100, ahead of Gemini at 79.
- Most usedChatGPTChatGPT sends 80,5% of AI chatbot referral traffic worldwide in setembro de 2026, ahead of Gemini at 10,9%.
- Best for codingClaudeClaude has the top rating in the LMArena WebDev arena among the chatbots, 1813 for claude-opus-5.5-max, ahead of ChatGPT at 1786. See the leaderboard.
- Best for research and current eventsChatGPTChatGPT leads the LMArena search arena among the chatbots, 1256 for gpt-5.6-sol-xhigh, ahead of Claude at 1229. In the Essentras AI Answer Index, Grok cites the most sources, 56.8 per answer over the last 90 days. See the leaderboard.
- Best for writingGeminiGemini has the top rating in the LMArena creative-writing category among the chatbots, 1519 for gemini-4-argon-high, ahead of Claude at 1517. See the leaderboard.
- Best open-weight optionkimi-k3-maxkimi-k3-max from Moonshot AI is the highest-rated open-license model on the LMArena text arena: 1488, rank 16, license Kimi K3 license. The top model, gemini-4-argon-high, rates 1525, a gap of 37 points. This is a model you can run yourself, not a consumer app. See the leaderboard.
- Best in your languageThe top model for each language in LMArena text-arena language categories (board of 8 de outubro de 2026). Open a language for its full leaderboard.
- francês: muse-spark-1.2 (xHigh) (Meta AI), 1534
- alemão: gemini-3-pro (Gemini), 1521
- espanhol: muse-spark-1.2 (xHigh) (Meta AI), 1505
- chinês: gemini-4-argon-high (Gemini), 1588
- japonês: claude-fable-5.1-max (Claude), 1534
- coreano: claude-fable-5.1-max (Claude), 1514
- russo: gemini-4-argon-high (Gemini), 1529
- polonês: claude-fable-5.1-max (Claude), 1510
ChatGPT: usage, quality and strengths
ChatGPT drives 80,5% of AI chatbot referral traffic worldwide in setembro de 2026, rank 1 of 6 chatbots StatCounter reports. That is -0,7 percentage points versus 12 months earlier. Its best model on the LMArena text arena is gpt-5.6-sol-xhigh, rated 1485 and ranked 20 of 414 models (board of 8 de outubro de 2026).
Rank among the chatbots with data (ratings are in the table above): quality 4 of 6, web search 1 of 5, coding 2 of 6, creative writing 3 of 6, citations per answer 3 of 6.
- ChatGPT: 900 mi usuários ativos semanais as of 27 de fevereiro de 2026. OpenAI, "Scaling AI for everyone"
- ChatGPT: 50 mi assinantes pagantes as of 27 de fevereiro de 2026. OpenAI, "Scaling AI for everyone"
ChatGPT usage share historyLMArena text leaderboardWebDev leaderboardSearch leaderboard
Gemini: usage, quality and strengths
Gemini drives 10,9% of AI chatbot referral traffic worldwide in setembro de 2026, rank 2 of 6 chatbots StatCounter reports. That is +8,1 percentage points versus 12 months earlier. Its best model on the LMArena text arena is gemini-4-argon-high, rated 1525 and ranked 1 of 414 models (board of 8 de outubro de 2026).
Rank among the chatbots with data (ratings are in the table above): quality 1 of 6, web search 3 of 5, coding 3 of 6, creative writing 1 of 6, citations per answer 6 of 6.
- Gemini: 950 mi usuários ativos mensais as of 22 de julho de 2026. Google, Alphabet Q2 2026 earnings remarks (Sundar Pichai)
Gemini usage share historyLMArena text leaderboardWebDev leaderboardSearch leaderboard
Claude: usage, quality and strengths
Claude drives 1,0% of AI chatbot referral traffic worldwide in setembro de 2026, rank 5 of 6 chatbots StatCounter reports. That is 0,0 percentage points versus 12 months earlier. Its best model on the LMArena text arena is claude-opus-5.5-high, rated 1507 and ranked 2 of 414 models (board of 8 de outubro de 2026).
Rank among the chatbots with data (ratings are in the table above): quality 2 of 6, web search 2 of 5, coding 1 of 6, creative writing 2 of 6, citations per answer 4 of 6.
No official user numbers for Claude are on file here.
Claude usage share historyLMArena text leaderboardWebDev leaderboardSearch leaderboard
Perplexity: usage, quality and strengths
Perplexity drives 5,8% of AI chatbot referral traffic worldwide in setembro de 2026, rank 3 of 6 chatbots StatCounter reports. That is -5,0 percentage points versus 12 months earlier. LMArena lists no Perplexity model among the top 150 models of its text arena.
Rank among the chatbots with data (ratings are in the table above): web search 5 of 5, citations per answer 2 of 6.
No official user numbers for Perplexity are on file here.
Perplexity usage share historyLMArena text leaderboardSearch leaderboard
Meta AI: usage, quality and strengths
StatCounter does not report Meta AI as a separate chatbot, so it has no usage figure here. Its best model on the LMArena text arena is muse-spark-1.3-max, rated 1494 and ranked 9 of 414 models (board of 8 de outubro de 2026).
Rank among the chatbots with data (ratings are in the table above): quality 3 of 6, coding 4 of 6, creative writing 4 of 6.
- Meta AI: 1 bi usuários ativos mensais as of 28 de maio de 2025. Meta annual shareholder meeting (Mark Zuckerberg), reported by TechCrunch
Grok: usage, quality and strengths
StatCounter does not report Grok as a separate chatbot, so it has no usage figure here. Its best model on the LMArena text arena is grok-4.20-beta1, rated 1475 and ranked 37 of 414 models (board of 8 de outubro de 2026).
Rank among the chatbots with data (ratings are in the table above): quality 5 of 6, web search 4 of 5, coding 5 of 6, creative writing 5 of 6, citations per answer 1 of 6.
No official user numbers for Grok are on file here.
LMArena text leaderboardWebDev leaderboardSearch leaderboard
DeepSeek: usage, quality and strengths
DeepSeek drives 0,0% of AI chatbot referral traffic worldwide in setembro de 2026, rank 6 of 6 chatbots StatCounter reports. That is -0,2 percentage points versus 12 months earlier. Its best model on the LMArena text arena is deepseek-v4.1-flash-max, rated 1475 and ranked 39 of 414 models (board of 8 de outubro de 2026).
Rank among the chatbots with data (ratings are in the table above): quality 6 of 6, coding 6 of 6, creative writing 6 of 6, citations per answer 5 of 6.
No official user numbers for DeepSeek are on file here.
DeepSeek usage share historyLMArena text leaderboardWebDev leaderboard
Microsoft Copilot: usage, quality and strengths
Microsoft Copilot drives 1,8% of AI chatbot referral traffic worldwide in setembro de 2026, rank 4 of 6 chatbots StatCounter reports. That is -2,3 percentage points versus 12 months earlier. We could not verify at a Microsoft page which models Copilot runs, so no quality rating is assigned.
- Microsoft 365 Copilot: 30 mi licenças pagas as of 29 de julho de 2026. Microsoft FY26 Q4 earnings (Satya Nadella)
Microsoft Copilot usage share historyLMArena text leaderboard
Free vs paid
Plan facts below were checked on each vendor's own plans page on 13 de outubro de 2026. A vendor whose page we could not read is not listed.
| Chatbot | Plans | What we checked | Source |
|---|---|---|---|
| Claude | Free ($0), Pro ($17 per month billed annually as $200 up front, or $20 billed monthly), Max (from $100 per month) | The page lists a Free plan at $0 and the Pro and Max prices shown. Checked 13 de outubro de 2026. | claude.com/pricing |
| Gemini | A free Gemini app tier, then Google AI Plus, Google AI Pro and Google AI Ultra | The page shows the free tier at no cost and lists the three paid plans. Consumer prices are shown in the visitor's local currency, so no single price is quoted. Checked 13 de outubro de 2026. | gemini.google/subscriptions |
| Perplexity | Standard (Free), Perplexity Pro, Perplexity Max | The page names the Standard (Free) plan and says it has no access to advanced AI models. It prints no price for Pro or Max, so none is quoted. Checked 13 de outubro de 2026. | Perplexity help center, subscription plans |
Not listed because we could not verify the plan details on the vendor's own page: ChatGPT (pricing pages block automated reads), Microsoft Copilot (no free tier stated on the page we could read), DeepSeek (only API pricing is published), Grok and Meta AI.
FAQ
Best AI chatbot: common questions
What is the best AI chatbot right now?▾
ChatGPT ranks first on our composite score, 94 out of 100 (setembro de 2026 usage data). It drives 80,5% of AI chatbot referral traffic worldwide (rank 1). Its best model, gpt-5.6-sol-xhigh, has an LMArena blind-vote rating of 1485. Gemini is second with 79.
What is the best free AI chatbot?▾
Among the chatbots whose free plans we verified on the vendor's own page on 13 de outubro de 2026 (Claude, Gemini, Perplexity), Gemini ranks highest on our composite, with 79 out of 100. We did not verify free plans for the other chatbots, so this is not a claim about them.
Which AI chatbot is the smartest?▾
By LMArena blind votes (board of 8 de outubro de 2026), Gemini has the highest-rated model of the listed chatbots: gemini-4-argon-high, rated 1525. Claude is next at 1507. The gap of 18 points is outside the overlap of their confidence intervals.
What is the most popular AI chatbot?▾
ChatGPT, with 80,5% of AI chatbot referral traffic worldwide in setembro de 2026 (StatCounter), ahead of Gemini at 10,9%. This measures traffic sent to websites, not the number of users.
Which is the best AI for coding?▾
In the LMArena WebDev arena, Claude has the highest-rated model of the listed chatbots: claude-opus-5.5-max, rated 1813, ahead of ChatGPT at 1786.
Which AI chatbot is best for research?▾
In the LMArena search arena, ChatGPT has the highest-rated model of the listed chatbots: gpt-5.6-sol-xhigh, rated 1256. In the Essentras AI Answer Index, Grok cites the most sources, 56.8 per answer over the last 90 days.
Is ChatGPT still the best?▾
Yes on our data: ChatGPT ranks 1 of 7 with 94 out of 100, it is rank 1 for usage (80,5% in setembro de 2026) and rank 4 of 6 on model quality.
How did you rank them?▾
Each chatbot gets a score out of 100 from four signals: model quality (40%, LMArena text arena), usage (30%, StatCounter referral share), web search (15%, LMArena search arena) and coding (15%, LMArena WebDev arena). Ratings are converted to each model's chance of beating the leader in a blind vote and usage is scored on a log scale, missing signals count at the lowest observed score, and a chatbot needs at least 2 signals to be ranked.
See what these chatbots say about your brand
Ask ChatGPT, Gemini, Claude, Perplexity and more the questions your buyers ask, free.