Links marked with * are affiliate links. If a purchase is made through such links, we receive a commission.
Look up how many people use ChatGPT, and you get three numbers.
One says more than 900 million, another about 1.2 billion, a third about 5.9 billion.
So which one is right? The most recent figure comes from OpenAI itself. On October 5, 2026, it said ChatGPT reaches about 1.2 billion people per week.
The figure of more than 900 million is from February 2026. The roughly 5.9 billion are website visits to chatgpt.com estimated by Similarweb. Similarweb labels them as covering the past three months, while Forbes cites the same figure as the September value.
Here you'll find the key ChatGPT statistics, with numbers and facts on users, traffic, subscribers, and usage. Company revenue and valuation are covered in the OpenAI statistics article. The four most important points first:
TL;DRKey Takeaways
On October 5, 2026, OpenAI said ChatGPT reaches about 1.2 billion people per week.
For February 2026, OpenAI reported more than 900 million weekly active users, more than 50 million consumer subscribers, and, separately, more than 9 million paying business users.
Similarweb lists about 5.9 billion website visits to chatgpt.com, which its own page labels as the past three months through September 2026.
According to OpenAI Signals v3.0, only 32% of ChatGPT messages were work-related in September 2026, down from 52% in July 2024. Nearly half of all messages are search or advice questions.
1. How many people use ChatGPT?
On December 5, 2022, five days after ChatGPT launched, Sam Altman wrote that it had reached more than one million users. In a July 2025 analysis, OpenAI repeated the milestones of one million users in five days and 100 million in two months.
According to an OpenAI and NBER study, more than 700 million people used ChatGPT every week in July 2025. By October 2025, it was more than 800 million, and by February 2026, more than 900 million.
OpenAI reported more than 100 million weekly ChatGPT users in India in February 2026.
2. How many monthly visits does ChatGPT get on the web?
The chart shows estimated monthly desktop and mobile web visits to ChatGPT's website. Its address changed from chat.openai.com to chatgpt.com. The figures are Similarweb estimates.
At launch in November 2022, Similarweb recorded about 153,000 visits on the first day and 266 million for December. In its report at the time, Similarweb put May 2023 at 1.8 billion visits.
The estimate then dipped over the summer to 1.5 billion in September. A later Similarweb report described the 2023 peak as 1.9 billion. September 2024 reached 3.1 billion. A PLOS One publication and the Australian ACCC report cited 3.8 billion visits in November and 3.7 billion in December 2024.
The series ends in August 2026 at 5.6 billion visits (Forbes citing Similarweb). The three-month figure of 5.9 billion cannot be inserted into the series as a monthly value.
3. Where does ChatGPT website traffic come from?
About a fifth of estimated desktop visits to chatgpt.com come from the United States. For Germany, Bitkom Research adds survey data.
3.1 Desktop visits by country
For September 2026, Similarweb listed the United States as the largest source of desktop visits to chatgpt.com. The percentages below are estimated desktop visits.
For Germany, Bitkom Research surveyed AI users. Of the 1,003 people aged 16 and older who were interviewed, 58% use AI at least occasionally, which is 579 respondents in the report. Of those, 71% said they use ChatGPT (multiple answers allowed).
ChatGPT
71%
Gemini
50%
Microsoft Copilot
43%
Meta AI
35%
DeepSeek
8%
Perplexity
7%
Grok
6%
Claude
5%
Source: Bitkom Research 2026
|
CC BY 4.0
gradually.ai
4. In which countries is ChatGPT available?
The map follows OpenAI's availability list from October 7, 2026. OpenAI lists 208 country, territory and regional entries.
ChatGPT availability according to OpenAI
0Available
0Not available
0Restricted
0Unknown
Available
Not available
Restricted
Unknown
Source: OpenAI, checked October 7, 2026
|
CC BY 4.0
gradually.ai
The simplified map shows 161 supported IDs. OpenAI lists Ukraine with regional exceptions that cannot be drawn individually here.
The new models are available in the API and in ChatGPT Work and Codex (rolling out on Plus, Pro, Business, Enterprise, and Edu), but not in regular ChatGPT Chat. GPT-6.2 Sol is therefore not automatically the newest model in every ChatGPT interface. For a broader overview, see All ChatGPT versions.
The chart shows the US list prices of the personal plans and a Business seat billed monthly. With annual billing, OpenAI lists a Business seat at $20 per month.
A model's knowledge cutoff marks the end of the training data represented in the model. The chart shows selected GPT generations and links to OpenAI documentation.
Among the GPT-6 models shown, GPT-6.2 Sol has the latest cutoff, June 1, 2026, as does GPT-6.1 Terra, which is not in the chart. GPT-6.2 Astra and GPT-6 Luna go through May 18, GPT-6.1 Sol and GPT-6 Astra go through April 30, while GPT-6 Sol goes through April 20. Web search can retrieve current pages, but it does not change the model's training data.
7. Is there a fixed answer length?
There is no single character limit for every ChatGPT model and plan. OpenAI's release notes gave manually selected Thinking a 256,000-token context window on February 20, 2026, including up to 128,000 output tokens. OpenAI's current ChatGPT help also says usage limits depend on plan, model, and workspace.
8. What do OpenAI's usage policies prohibit?
OpenAI's usage policies prohibit, among other things, threats and harassment, promotion of self-harm, sexual violence, weapons development, child abuse, fraud, spam, and political election interference. They also prohibit automating consequential decisions in sensitive areas without human review.
The policies apply across OpenAI services and also cover privacy, protection of minors, and licensed professional advice. See the current OpenAI usage policies.
9. What do studies say about how people use ChatGPT?
On October 26, 2026, OpenAI released version 3.0 of its Signals data. From July 2024 through September 2026, it automatically classified 1 million messages from about 160 countries every month, 27 million in total. Only Free, Go, Plus, and Pro accounts are included, so neither Enterprise nor Codex.
Seeking information (facts, products, recipes) and practical guidance (how-to, tutoring, health, ideas) together account for 48% of all messages, up from 43% in July 2024. Writing fell from 34% to 21% over the same period.
9.1 Personal use grows faster than work
OpenAI tags every message as work-related or personal. The work share fell from 52% in July 2024 to 32% in September 2026. OpenAI only publishes shares, so work messages could still have grown in absolute numbers.
9.3 Almost every other message is a search or advice question
Seeking information and practical guidance make up 53% of personal messages but only 38% of work messages. Because the personal share grew from 48% to 68%, the overall figure rose too. Most of the increase from 43% to 48% comes from this shift.
Seeking information
Practical guidance
Share of messages
All messages
48.3%Seeking information 20.2% · Practical guidance 28.1%
Personal messages
53.0%Seeking information 22.5% · Practical guidance 30.5%
Work messages
38.1%Seeking information 15.2% · Practical guidance 22.9%
9.4 At work, people ask more about products and prices
For the US, OpenAI maps work messages to 165 activities from the O*NET occupational database. Questions about products, services, and prices rose from 2.0% to 3.6% of work messages between Q1 2025 and Q3 2026.
Subject research declined, by about two thirds for technology and for historical and social issues. Programming also fell from 6.7% to 2.4%, though Codex is not part of the data.
9.5 Where ChatGPT is more common, people search more
In the quarter of countries with the highest ChatGPT use per capita, 24% of messages in Q3 2026 were information searches, compared with 16% in the quarter with the lowest use. In early 2025, both groups were still close.
Across about 150 countries, the rank correlation between use per capita and search share rose from 0.29 to 0.74. Germany climbed from 32nd to 19th place in use per capita.
An OpenAI study with Harvard researcher David Deming analyzed consumer-plan ChatGPT conversations. The topic distribution below uses about 1.1 million randomly sampled conversations from May 15, 2024, through June 26, 2025, classified automatically.
Practical guidance, seeking information, and writing are among the three most common topics. The study classifies about 70% of consumer messages as unrelated to work. In June 2025, writing accounted for 42% of work-related messages, the most common work-related use. About two-thirds of those writing messages asked ChatGPT to revise existing text.
9.7 ChatGPT use by age in the US
A separate Pew survey of 5,123 U.S. adults ran from February 24 to March 2, 2025. It asked whether respondents had ever used ChatGPT.
In August 2026, OpenAI reported up to 70 million ChatGPT conversations per week in which people tested their knowledge or requested more practice. In the United States, prompts about classwork and homework peaked above 460 million messages per week during the school year and remained above 180 million in the summer.
A separate Pew survey fielded from September 25 to October 9, 2025, found that 54% of U.S. teens had used AI chatbots for schoolwork. The survey named ChatGPT, Copilot, and Character.ai.
An earlier Pew survey of 1,391 U.S. teens ages 13 to 17 found that the share who said they had used ChatGPT for schoolwork rose from 13% in 2023 to 26% in 2024. The 2024 survey ran from September 18 to October 10.
10. How does ChatGPT compare with other chatbots?
Similarweb estimated that ChatGPT had the largest share of worldwide website traffic in its selection of AI chatbots in September 2026, ahead of Gemini and Claude.
OpenAI does not publish a ranking of answer quality by language, nor for programming languages. How well ChatGPT performs in individual languages is covered in What Languages Does ChatGPT Support? And How Well?
12. How much data and compute goes into training?
OpenAI's 2020 GPT-3 paper describes 570 gigabytes of filtered Common Crawl data, equivalent to roughly 400 billion byte-pair tokens. GPT-3 was trained on 300 billion tokens across all the datasets described in the paper. According to the same source, training took 3.14 × 10²³ floating-point operations (FLOP).
OpenAI no longer discloses these figures for later ChatGPT models. The GPT-4 technical report from March 2023 explicitly gives no details on model size, hardware, training compute or dataset.
Epoch AI estimates GPT-4's training compute at about 2.1 × 10²⁵ FLOP, roughly 67 times that of GPT-3. Epoch itself rates this estimate as speculative.
OpenAI is more specific about its open models. According to the model card from August 5, 2025, training gpt-oss-120b took 2.1 million H100 GPU hours. It was trained on a text-only dataset with “trillions of tokens”, focused on STEM, coding and general knowledge.
13. How much energy does a ChatGPT query use?
In June 2025, Sam Altman wrote that an average ChatGPT query used about 0.34 watt-hours and 0.000085 gallons of water. An independent Epoch AI estimate put a typical GPT-4o query with 500 output tokens at about 0.3 watt-hours in February 2025.
13.1 Company figures and estimates per query
In August 2025, Google reported 0.24 watt-hours and 0.26 milliliters of water for a median text prompt in the Gemini app, measured across the whole system including idle reserve capacity and data center overhead.
ChatGPT (average query, OpenAI statement, June 2025)
0.34 Wh
GPT-4o(Epoch AI estimate, 500 output tokens, Feb. 2025)
about 0.3 Wh
Gemini Apps (median text prompt, Google, May 2025)
A study by Jegham et al. estimates a short GPT-4o query with 100 input and 300 output tokens at about 0.42 watt-hours. The Jegham et al. estimate follows in the next chart, which compares several models and query lengths with one method.
Besides the model, the length of a query matters most. Epoch AI puts 10,000 input tokens at about 2.5 watt-hours and 100,000 input tokens at almost 40 watt-hours. For a long query with 10,000 input and 1,500 output tokens, Jegham et al. arrive at 2.9 watt-hours for GPT-4o and 12.2 watt-hours for the reasoning model o3.
OpenAI o3, long query
12.2 Wh
GPT-4o, long query
2.9 Wh
OpenAI o3, short query
1.2 Wh
GPT-4o, short query
0.42 Wh
Source: Jegham et al., November 2025
|
CC BY 4.0
gradually.ai
Reasoning adds to the bill. In Hugging Face's AI Energy Score, open reasoning models used 30 times more energy on average than models without reasoning in December 2025. With reasoning enabled, they produced 300 to 800 times more tokens.
Mistral published a lifecycle analysis in July 2025. It puts a 400-token response from its Le Chat assistant at 1.14 grams of CO₂ equivalent and 45 milliliters of water.
Newer figures exist only for OpenAI as a whole. According to leaked financial statements reported by Ars Technica in June 2026, OpenAI's operating loss grew from $8.78 billion in 2024 to $20.92 billion in 2025.
OpenAI plans for even higher spending in the coming years. According to Reuters, a company presentation reported by the Financial Times in September 2026 forecasts about $856 billion in spending on compute and infrastructure by the end of 2030. For 2026 to 2030 it projects $278 billion in cumulative negative free cash flow.
For OpenAI's revenue and funding, see the article on OpenAI statistics.
15. How many messages does ChatGPT process?
In December 2024, OpenAI reported about 1 billion messages per day. In July 2025, Axios reported more than 2.5 billion prompts per day, about 330 million of them from the US. An OpenAI spokesperson confirmed the figures, as Gigazine summarized on July 22, 2025.
The OpenAI and NBER study puts July 2025 at about 18 billion messages per week, or more than 2.5 billion per day. That confirms the order of magnitude.
OpenAI scores 97 out of 100 in the AI vendor popularity index, rank 1 of 39. The value combines search demand for the vendor and its AI brands, usage via OpenRouter and Hugging Face and mentions in tech media.
In the EU, OpenAI is at rank 1 of 36, in Germany at rank 1 of 36. Use the switch above the chart to change between world, EU and Germany.
Vendor popularity index
0 to 100, higher is more popular
OpenAI97
Google91
Anthropic88
DeepSeek85
Meta79
Alibaba78
Perplexity75
ElevenLabs73
MiniMax72
Z.ai72
Moonshot AI72
Microsoft71
Mistral AI70
xAI70
Stability AI69
ByteDance68
NVIDIA68
Tencent66
Black Forest Labs64
Cohere62
Liquid AI61
Skywork AI61
Kuaishou60
Xiaomi60
Amazon60
Runway60
StepFun60
IBM57
ShengShu Technology57
Mapika57
Baidu56
inclusionAI53
Upstage49
Bilibili Index Team48
Nace.AI47
AI2147
Arcee AI47
Institute of Foundation Models45
Nex AGI37
020406080100
OpenAI, OpenAI: 9797
Google, Google: 9191
Anthropic, Anthropic: 8888
DeepSeek, DeepSeek: 8585
Meta, Meta: 7979
Alibaba, Alibaba: 7878
Perplexity, Perplexity: 7575
ElevenLabs, ElevenLabs: 7373
MiniMax, MiniMax: 7272
Z.ai, Z.ai: 7272
Moonshot AI, Moonshot AI: 7272
Microsoft, Microsoft AI: 7171
Mistral AI, Mistral AI: 7070
xAI, xAI: 7070
Stability AI, Stability AI: 6969
ByteDance, ByteDance: 6868
NVIDIA, NVIDIA: 6868
Tencent, Tencent: 6666
Black Forest Labs, Black Forest Labs: 6464
Cohere, Cohere: 6262
Liquid AI, Liquid AI: 6161
Skywork AI, Skywork AI: 6161
Kuaishou, Kuaishou: 6060
Xiaomi, Xiaomi: 6060
Amazon, Amazon: 6060
Runway, Runway: 6060
StepFun, StepFun: 6060
IBM, IBM: 5757
ShengShu Technology, ShengShu Technology: 5757
Mapika, Mapika: 5757
Baidu, Baidu: 5656
inclusionAI, inclusionAI: 5353
Upstage, Upstage: 4949
Bilibili Index Team, Bilibili Index Team: 4848
Nace.AI, Nace.AI: 4747
AI21, AI21: 4747
Arcee AI, Arcee AI: 4747
Institute of Foundation Models, Institute of Foundation Models: 4545
Nex AGI, Nex AGI: 3737
OpenAI
Google
Anthropic
DeepSeek
Meta
Alibaba
Perplexity
ElevenLabs
MiniMax
Z.ai
Moonshot AI
Microsoft
Mistral AI
xAI
Stability AI
ByteDance
NVIDIA
Tencent
Black Forest Labs
Cohere
Liquid AI
Skywork AI
Kuaishou
Xiaomi
Amazon
Runway
StepFun
IBM
ShengShu Technology
Mapika
Baidu
inclusionAI
Upstage
Bilibili Index Team
Nace.AI
AI21
Arcee AI
Institute of Foundation M…
Nex AGI
As of Nov 22, 2026 · Source: gradually.ai popularity index
17. How popular are OpenAI's models?
OpenAI's most popular model is GPT-6.1 Terra with 87 out of 100 (rank 2 of 881 AI models with a popularity score). In total, 54 OpenAI models have a score; the chart highlights them.
In the EU, GPT-6 Astra is at rank 1 of 613, in Germany, GPT-6 Astra is at rank 1 of 692. Use the switch above the chart to change between world, EU and Germany.
Popularity of OpenAI models
Popularity index from 0 to 100, higher is more popular.
24 of 137 models
Claude Opus 5.588
GPT-6.1 Terra87
GLM-5.387
Kimi K386
Nano Banana 286
Claude Haiku 5.586
GPT-6 Luna86
Gemini 4 Argon85
GPT-6.1 Sol85
DeepSeek-V4.1-Flash85
GLM-5.3-Flash84
Sora 284
GPT-6 Astra83
MiniMax H383
Nano Banana Pro82
Veo 3.182
Qwen 3.8 27B82
DeepSeek V4.1 Flash 101781
GPT-581
GLM-5.281
GPT Image 280
GPT-5.6 Luna77
GPT Image 2.5 Sunburst75
GPT OSS 120B75
020406080100
Claude Opus 5.5, Anthropic: 8888
GPT-6.1 Terra, OpenAI: 8787
GLM-5.3, Z.ai: 8787
Kimi K3, Moonshot AI: 8686
Nano Banana 2, Google: 8686
Claude Haiku 5.5, Anthropic: 8686
GPT-6 Luna, OpenAI: 8686
Gemini 4 Argon, Google DeepMind: 8585
GPT-6.1 Sol, OpenAI: 8585
DeepSeek-V4.1-Flash, DeepSeek: 8585
GLM-5.3-Flash, Z.ai: 8484
Sora 2, OpenAI: 8484
GPT-6 Astra, OpenAI: 8383
MiniMax H3, MiniMax: 8383
Nano Banana Pro, Google: 8282
Veo 3.1, Google: 8282
Qwen 3.8 27B, Alibaba: 8282
DeepSeek V4.1 Flash 1017, DeepSeek: 8181
GPT-5, OpenAI: 8181
GLM-5.2, Z.ai: 8181
GPT Image 2, OpenAI: 8080
GPT-5.6 Luna, OpenAI: 7777
GPT Image 2.5 Sunburst, OpenAI: 7575
GPT OSS 120B, OpenAI: 7575
Claude Opus 5.5
GPT-6.1 Terra
GLM-5.3
Kimi K3
Nano Banana 2
Claude Haiku 5.5
GPT-6 Luna
Gemini 4 Argon
GPT-6.1 Sol
DeepSeek-V4.1-Flash
GLM-5.3-Flash
Sora 2
GPT-6 Astra
MiniMax H3
Nano Banana Pro
Veo 3.1
Qwen 3.8 27B
DeepSeek V4.1 Flash 1017
GPT-5
GLM-5.2
GPT Image 2
GPT-5.6 Luna
GPT Image 2.5 Sunburst
GPT OSS 120B
Source: gradually.ai popularity index
18. How good are OpenAI's models?
In the gradually.ai overall index, which combines independent benchmarks into one value, GPT-6.2 Astra is OpenAI's strongest model: rank 4 of 99 with 126.6 points. 20 OpenAI models are ranked in the index. 100 points is the average of a fixed reference group; 15 points more means a model wins about three of four benchmark comparisons.
Artificial Analysis uses AA-Omniscience to measure how often a model gives a wrong answer to a knowledge question instead of admitting it does not know. OpenAI's best result comes from GPT-6.1 Terra: 35% of its answers that are not correct are wrong guesses (rank 30 of 76, rank 1 hallucinates least). The median of all measured models is 40%.
10 OpenAI models have a measurement. Each model counts with its best measured setting. The rate describes reliability, not safety in the sense of jailbreaks or harmful content.
Hallucination rate
Share of wrong answers among all answers that are not correct, lower is better. Best measured setting per model.
GPT OSS 120B is the fastest with 145 output tokens per second. The median of all measured models is 72 tokens per second. OpenRouter has measured 22 OpenAI models.
Speed
27 of 100 models
Nemotron 3.5 Lightning304tokens/s · Median across providers on OpenRouter from the readings of the last 24 hours (each covering the last 30 minutes), excluding time to first token · Retrieved 2026-11-22
MiniMax M2.7203tokens/s · Median across providers on OpenRouter from the readings of the last 24 hours (each covering the last 30 minutes), excluding time to first token · Retrieved 2026-11-22
Claude Haiku 5.5162tokens/s · Median across providers on OpenRouter from the readings of the last 24 hours (each covering the last 30 minutes), excluding time to first token · Retrieved 2026-11-22
DeepSeek-V4.1-Flash159tokens/s · Median across providers on OpenRouter from the readings of the last 24 hours (each covering the last 30 minutes), excluding time to first token · Retrieved 2026-11-22
Muse Spark 1.3150tokens/s · Median across providers on OpenRouter from the readings of the last 24 hours (each covering the last 30 minutes), excluding time to first token · Retrieved 2026-11-22
GPT OSS 120B145tokens/s · Median across providers on OpenRouter from the readings of the last 24 hours (each covering the last 30 minutes), excluding time to first token · Retrieved 2026-11-22
DeepSeek V4.1 Flash 1017144tokens/s · Median across providers on OpenRouter from the readings of the last 24 hours (each covering the last 30 minutes), excluding time to first token · Retrieved 2026-11-22
Gemini 3.5 Flash-Lite138tokens/s · Median across providers on OpenRouter from the readings of the last 24 hours (each covering the last 30 minutes), excluding time to first token · Retrieved 2026-11-22
Gemini 3.5 Flash138tokens/s · Median across providers on OpenRouter from the readings of the last 24 hours (each covering the last 30 minutes), excluding time to first token · Retrieved 2026-11-22
Nemotron 3 Ultra136tokens/s · Median across providers on OpenRouter from the readings of the last 24 hours (each covering the last 30 minutes), excluding time to first token · Retrieved 2026-11-22
Gemini 3.1 Flash-Lite134tokens/s · Median across providers on OpenRouter from the readings of the last 24 hours (each covering the last 30 minutes), excluding time to first token · Retrieved 2026-11-22
Inkling Small126tokens/s · Median across providers on OpenRouter from the readings of the last 24 hours (each covering the last 30 minutes), excluding time to first token · Retrieved 2026-11-22
GPT-5.6 Luna118tokens/s · Median across providers on OpenRouter from the readings of the last 24 hours (each covering the last 30 minutes), excluding time to first token · Retrieved 2026-11-22
Mistral Small 4115tokens/s · Median across providers on OpenRouter from the readings of the last 24 hours (each covering the last 30 minutes), excluding time to first token · Retrieved 2026-11-22
Gemini 3.6 Flash114tokens/s · Median across providers on OpenRouter from the readings of the last 24 hours (each covering the last 30 minutes), excluding time to first token · Retrieved 2026-11-22
Grok 4.20 Reasoning109tokens/s · Median across providers on OpenRouter from the readings of the last 24 hours (each covering the last 30 minutes), excluding time to first token · Retrieved 2026-11-22
Inkling107tokens/s · Median across providers on OpenRouter from the readings of the last 24 hours (each covering the last 30 minutes), excluding time to first token · Retrieved 2026-11-22
Claude Sonnet 5.5106tokens/s · Median across providers on OpenRouter from the readings of the last 24 hours (each covering the last 30 minutes), excluding time to first token · Retrieved 2026-11-22
Mistral Medium 3.5106tokens/s · Median across providers on OpenRouter from the readings of the last 24 hours (each covering the last 30 minutes), excluding time to first token · Retrieved 2026-11-22
GPT OSS 20B105tokens/s · Median across providers on OpenRouter from the readings of the last 24 hours (each covering the last 30 minutes), excluding time to first token · Retrieved 2026-11-22
GPT-5 nano104tokens/s · Median across providers on OpenRouter from the readings of the last 24 hours (each covering the last 30 minutes), excluding time to first token · Retrieved 2026-11-22
GPT-4.1 nano99tokens/s · Median across providers on OpenRouter from the readings of the last 24 hours (each covering the last 30 minutes), excluding time to first token · Retrieved 2026-11-22
o398tokens/s · Median across providers on OpenRouter from the readings of the last 24 hours (each covering the last 30 minutes), excluding time to first token · Retrieved 2026-11-22
GPT-5.4 mini92tokens/s · Median across providers on OpenRouter from the readings of the last 24 hours (each covering the last 30 minutes), excluding time to first token · Retrieved 2026-11-22
GPT-5 mini87tokens/s · Median across providers on OpenRouter from the readings of the last 24 hours (each covering the last 30 minutes), excluding time to first token · Retrieved 2026-11-22
GPT-4.1 mini84tokens/s · Median across providers on OpenRouter from the readings of the last 24 hours (each covering the last 30 minutes), excluding time to first token · Retrieved 2026-11-22
GPT-6 Sol79tokens/s · Median across providers on OpenRouter from the readings of the last 24 hours (each covering the last 30 minutes), excluding time to first token · Retrieved 2026-11-22
tokens/s0100200300400
Nemotron 3.5 Lightning, NVIDIA: 304 tokens per second · Median across providers on OpenRouter from the readings of the last 24 hours (each covering the last 30 minutes), excluding time to first token · Retrieved 2026-11-22304
MiniMax M2.7, MiniMax: 203 tokens per second · Median across providers on OpenRouter from the readings of the last 24 hours (each covering the last 30 minutes), excluding time to first token · Retrieved 2026-11-22203
Claude Haiku 5.5, Anthropic: 162 tokens per second · Median across providers on OpenRouter from the readings of the last 24 hours (each covering the last 30 minutes), excluding time to first token · Retrieved 2026-11-22162
DeepSeek-V4.1-Flash, DeepSeek: 159 tokens per second · Median across providers on OpenRouter from the readings of the last 24 hours (each covering the last 30 minutes), excluding time to first token · Retrieved 2026-11-22159
Muse Spark 1.3, Meta: 150 tokens per second · Median across providers on OpenRouter from the readings of the last 24 hours (each covering the last 30 minutes), excluding time to first token · Retrieved 2026-11-22150
GPT OSS 120B, OpenAI: 145 tokens per second · Median across providers on OpenRouter from the readings of the last 24 hours (each covering the last 30 minutes), excluding time to first token · Retrieved 2026-11-22145
DeepSeek V4.1 Flash 1017, DeepSeek: 144 tokens per second · Median across providers on OpenRouter from the readings of the last 24 hours (each covering the last 30 minutes), excluding time to first token · Retrieved 2026-11-22144
Gemini 3.5 Flash-Lite, Google: 138 tokens per second · Median across providers on OpenRouter from the readings of the last 24 hours (each covering the last 30 minutes), excluding time to first token · Retrieved 2026-11-22138
Gemini 3.5 Flash, Google: 138 tokens per second · Median across providers on OpenRouter from the readings of the last 24 hours (each covering the last 30 minutes), excluding time to first token · Retrieved 2026-11-22138
Nemotron 3 Ultra, NVIDIA: 136 tokens per second · Median across providers on OpenRouter from the readings of the last 24 hours (each covering the last 30 minutes), excluding time to first token · Retrieved 2026-11-22136
Gemini 3.1 Flash-Lite, Google: 134 tokens per second · Median across providers on OpenRouter from the readings of the last 24 hours (each covering the last 30 minutes), excluding time to first token · Retrieved 2026-11-22134
Inkling Small, Thinking Machines Lab: 126 tokens per second · Median across providers on OpenRouter from the readings of the last 24 hours (each covering the last 30 minutes), excluding time to first token · Retrieved 2026-11-22126
GPT-5.6 Luna, OpenAI: 118 tokens per second · Median across providers on OpenRouter from the readings of the last 24 hours (each covering the last 30 minutes), excluding time to first token · Retrieved 2026-11-22118
Mistral Small 4, Mistral AI: 115 tokens per second · Median across providers on OpenRouter from the readings of the last 24 hours (each covering the last 30 minutes), excluding time to first token · Retrieved 2026-11-22115
Gemini 3.6 Flash, Google: 114 tokens per second · Median across providers on OpenRouter from the readings of the last 24 hours (each covering the last 30 minutes), excluding time to first token · Retrieved 2026-11-22114
Grok 4.20 Reasoning, xAI: 109 tokens per second · Median across providers on OpenRouter from the readings of the last 24 hours (each covering the last 30 minutes), excluding time to first token · Retrieved 2026-11-22109
Inkling, Thinking Machines Lab: 107 tokens per second · Median across providers on OpenRouter from the readings of the last 24 hours (each covering the last 30 minutes), excluding time to first token · Retrieved 2026-11-22107
Claude Sonnet 5.5, Anthropic: 106 tokens per second · Median across providers on OpenRouter from the readings of the last 24 hours (each covering the last 30 minutes), excluding time to first token · Retrieved 2026-11-22106
Mistral Medium 3.5, Mistral AI: 106 tokens per second · Median across providers on OpenRouter from the readings of the last 24 hours (each covering the last 30 minutes), excluding time to first token · Retrieved 2026-11-22106
GPT OSS 20B, OpenAI: 105 tokens per second · Median across providers on OpenRouter from the readings of the last 24 hours (each covering the last 30 minutes), excluding time to first token · Retrieved 2026-11-22105
GPT-5 nano, OpenAI: 104 tokens per second · Median across providers on OpenRouter from the readings of the last 24 hours (each covering the last 30 minutes), excluding time to first token · Retrieved 2026-11-22104
GPT-4.1 nano, OpenAI: 99 tokens per second · Median across providers on OpenRouter from the readings of the last 24 hours (each covering the last 30 minutes), excluding time to first token · Retrieved 2026-11-2299
o3, OpenAI: 98 tokens per second · Median across providers on OpenRouter from the readings of the last 24 hours (each covering the last 30 minutes), excluding time to first token · Retrieved 2026-11-2298
GPT-5.4 mini, OpenAI: 92 tokens per second · Median across providers on OpenRouter from the readings of the last 24 hours (each covering the last 30 minutes), excluding time to first token · Retrieved 2026-11-2292
GPT-5 mini, OpenAI: 87 tokens per second · Median across providers on OpenRouter from the readings of the last 24 hours (each covering the last 30 minutes), excluding time to first token · Retrieved 2026-11-2287
GPT-4.1 mini, OpenAI: 84 tokens per second · Median across providers on OpenRouter from the readings of the last 24 hours (each covering the last 30 minutes), excluding time to first token · Retrieved 2026-11-2284
GPT-6 Sol, OpenAI: 79 tokens per second · Median across providers on OpenRouter from the readings of the last 24 hours (each covering the last 30 minutes), excluding time to first token · Retrieved 2026-11-2279
Via the API, OpenAI's cheapest model, GPT OSS 20B, costs $0.06 per 1M tokens. The most expensive is GPT-6 Astra at $20.00. Both are blended prices of three parts input to one part output, as in the LLM price index.
API prices
Blended price in US dollars per 1M tokens, 3 parts input to 1 part output
Frequently asked questions about ChatGPT statistics
About 1.2 billion people each week, OpenAI said on October 5, 2026.
More than 50 million consumer subscribers and, separately, more than 9 million paying business users, OpenAI reported on February 27, 2026.
About 5.6 billion in August 2026, the last monthly value of Similarweb's estimate for chatgpt.com. For September, Similarweb shows about 5.9 billion but labels it as the past three months, while Forbes cites it as a monthly value. The sources disagree here.
By desktop visits, the United States. Similarweb attributes 21.07% of estimated desktop visits to chatgpt.com in September 2026 to it. OpenAI reported more than 100 million weekly ChatGPT users in India in February 2026.
OpenAI lists 59 languages for the ChatGPT interface. For a language-by-language overview, see our ChatGPT language guide.
GPT-6.2 Sol and GPT-6.2 Astra (released November 17, 2026) are OpenAI's newest GPT models for the API. They are also available in ChatGPT Work and Codex, but not the Chat tab. GPT-6 is still rolling out in Chat to eligible paid users, and the release notes name GPT-6 Luna for Free and Go. The current model guide still says GPT-5.6 Luna is the default, so check your account's model picker. For a model-by-model history, see our ChatGPT versions overview.
Seeking information and practical guidance. Together they made up 48% of all messages in September 2026, according to OpenAI Signals v3.0, while writing accounted for 21%. Only 32% of messages were work-related.
It depends on the model. In OpenAI's model registry, the latest listed GPT-6 cutoff is June 1, 2026, for GPT-6.2 Sol, tied with GPT-6.1 Terra. Web search can retrieve current sources, but it does not change the model's training data.
Finn Hillebrandt is the founder of Gradually AI, an SEO and AI expert. He helps online entrepreneurs simplify and automate their processes and marketing with AI. Finn shares his knowledge here on the blog in 50+ articles as well as through the AI Business Club.