Claude vs ChatGPT Statistics 2026: Users, Cost and Share

Anthropic earns an estimated 40% of enterprise LLM spend against OpenAI’s 27%. That reverses the 50% share OpenAI held in 2023, though consumer scale runs the other way. OpenAI publishes country-level data covering more than 1 billion people putting ChatGPT to work, while no Anthropic source captured for this article publishes a Claude user total. Those two figures frame every Claude vs ChatGPT statistics comparison worth running, because the two companies measure different things on purpose.

Both companies’ own documentation, research, and launch posts supply the data below, alongside one enterprise spending survey. Coverage spans API pricing, context limits, competing benchmark claims, usage composition by country and topic, and the safety behavior separating the two assistants day to day.

Key Takeaways

• Anthropic’s estimated enterprise share climbed from 12% in 2023 to 40%, while OpenAI’s fell from 50% to 27%.
• In coding, Anthropic holds an estimated 54% market share against OpenAI’s 21%.
• Google increased its enterprise share from 7% in 2023 to 21% in 2025, so the contest is three-way rather than a duel.
• Enterprise open-source share declined from 19% last year to 11% today.
• OpenAI charges $4.00 per million input tokens for GPT-5.6 Sol on short context and $8.00 on long context, while Anthropic’s model overview lists Claude Opus 5 at $5 per million input tokens with a 1 million token context window.
• Anthropic reduced biology-related fallbacks by about 85%, but Claude Fable 5 still falls back to Opus 5 on dual-use requests.
• Chinese open-source models account for just 1% of total LLM API usage, roughly 10% of enterprise open-source.

Editor’s Choice

• Foundation model APIs absorbed $12.5 billion of enterprise spend in 2025.
• The infrastructure layer captured $18 billion in 2025.
• Claude Fable 5 is priced at $10 per million input tokens and $50 per million output tokens.
• GPT-5.6 Sol is priced at $4 per million input tokens and $20 per million output tokens.
• OpenAI reports GPT-5.6 Sol setting a new high of 53.6 on Agents’ Last Exam, eclipsing Claude Fable 5 by 13.1 points.
• Anthropic reports Claude Opus 4.6 leading GPT-5.2 by around 144 Elo points on GDPval-AA.
• Anthropic’s classifier identified 93% of Claude conversations as producing an artifact.

Claude vs ChatGPT Statistics: The Enterprise Spending Split

• Anthropic, OpenAI and Google together account for 88% of enterprise LLM API usage.
• The remaining 12% is spread across Meta’s Llama, Cohere, Mistral and a long tail of smaller providers.
• Anthropic’s enterprise share stood at 24% the year before.
• Anthropic commands an estimated 54% of the coding market, compared to 21% for OpenAI.
• That coding share is up from 42% just six months ago.
• Anthropic has held 18 months atop the LLM leaderboards for coding, starting with Claude Sonnet 3.5 in June 2024.
• OpenAI lost nearly half of its enterprise share over the same period.
• Llama remains the most widely adopted open-weight model in the enterprise despite falling off the frontier pace.

Source: Menlo Ventures, 2025: The State of Generative AI in the Enterprise, December 2025

Anthropic’s share therefore moved 28 percentage points in two years. Neither firm publishes these figures itself, which is exactly why the survey deserves naming in prose rather than burial in a footnote. Google Gemini tripled its slice over the same period. Any two-horse framing understates how contested the category has become.

About This Data

Thirteen sources captured in August 2026 supply the Claude vs ChatGPT statistics on this page. Twelve are first-party publications from Anthropic and OpenAI, covering model documentation, API pricing pages, launch posts and research reports. One enterprise spending survey from Menlo Ventures completes the set. Only sources publishing their own data qualified, and figures refresh when those sources publish new editions rather than on a fixed schedule.

Don’t chase tech news. We track it for you.

One weekly briefing with the launches, AI developments, and breaches that matter. No filler.

Email address

Subscribe Now

Where the 2025 Enterprise AI Budget Actually Went

• The infrastructure layer was up 2.0x from $9.2 billion in 2024.
• Foundation model APIs power the intelligence behind all AI applications at $12.5 billion.
• Model training infrastructure took $4.0 billion, enabling frontier labs and enterprises to train and adapt models.
• AI infrastructure took $1.5 billion, managing the storage, retrieval, and orchestration of data that connects LLMs to enterprise systems.
• Infrastructure represented the other half of all generative AI spending in 2025.
• Enterprises remain particularly cautious toward Chinese open-source models despite their progress and growing popularity among startups.

By the numbers: Foundation model APIs absorbed $12.5 billion of the $18 billion enterprise AI infrastructure layer in 2025, according to Menlo Ventures. That single line item settles the Anthropic and OpenAI share contest, since it is the category both sell into most directly.

Recent Developments

• June 30, 2026: Anthropic introduced Claude Sonnet 5, delivering frontier performance across coding, agents, and professional work at scale.
• July 24, 2026: Anthropic introduced Claude Opus 5, a step-change improvement for the Opus tier powering long-running agents.
• July 27, 2026: Cognizant and Anthropic expanded their partnership to bring Claude to enterprise clients.
• July 30, 2026: OpenAI reduced the price of GPT-5.6 Luna by 80% and also cut GPT-5.6 Terra pricing.
• August 2, 2026: The EU began requiring AI providers serving its market to mark AI-generated content.
• August 7, 2026: Anthropic cut biology-related fallbacks on Claude Fable 5 by about 85%.
• August 21, 2026: OpenAI dropped the API and credit pricing of GPT-5.6 Sol by over 20% for the next 3 months.

Claude and ChatGPT API Prices per Million Tokens

• Claude Opus 5 is priced at $5 per million input tokens and $25 per million output tokens.
• Claude Sonnet 5 is priced at $2 per million input tokens and $10 per million output tokens.
• Claude Haiku 4.5 is priced at $1 per million input tokens and $5 per million output tokens.
• GPT-5.6 Terra is priced at $2 per million input tokens and $12 per million output tokens.
• GPT-5.6 Luna is priced at $0.20 per million input tokens and $1.20 per million output tokens.
• Batch processing takes GPT-5.6 Sol to $2.00 input and $10.00 output per million tokens.
• Cached input on GPT-5.6 Sol costs $0.40 per million tokens on short context.
• Cached input on GPT-5.6 Luna costs $0.02 per million tokens on short context.
• Claude Fable 5 is described as next-generation intelligence for long-running agents.

Anthropic’s cheapest output tier costs roughly 4 times more per million tokens than OpenAI’s. That reshapes the arithmetic for anyone running high-volume classification rather than agentic coding. OpenAI has been cutting low-end prices repeatedly through 2026, and the gap has widened rather than closed.

Model Specifications Side by Side

• Claude Fable 5, Opus 5, and Sonnet 5 each carry a 1 million token context window.
• Claude Haiku 4.5 carries a 200,000-token context window.
• Every GPT-5.6 model carries a context window of 1.05 million tokens.
• Max output runs to 128,000 tokens on Fable 5, Opus 5, and Sonnet 5, and 64,000 tokens on Haiku 4.5.
• Max output on every GPT-5.6 model is 128,000 tokens.
• Claude Opus 5 carries a reliable knowledge cutoff of May 2026, the most recent in the Claude lineup.
• All three GPT-5.6 models share a knowledge cutoff of February 16, 2026.
• Thinking is adaptive and always on for Fable 5, and adaptive for Opus 5 and Sonnet 5.

Source: Anthropic and OpenAI developer documentation, August 2026

Both vendors document these specifications publicly, in Anthropic’s model overview and OpenAI’s model reference.

What can Claude do that ChatGPT can’t?

On documented specifications, very little separates them at the top end. Claude Opus 5 carries a reliable knowledge cutoff of May 2026. That sits later than the GPT-5.6 family, and Anthropic’s model overview lists a single input price for that full window, while OpenAI counters with a marginally larger window and far cheaper entry tiers.

OpenAI’s Long-Context Surcharge and Anthropic’s Single Listed Rate

• GPT-5.6 Sol input pricing rises from $4.00 to $8.00 per million tokens on long context.
• GPT-5.6 Sol output pricing rises from $20.00 to $30.00 per million tokens on long context.
• GPT-5.6 Terra input pricing rises from $2.00 to $4.00 per million tokens.
• GPT-5.6 Luna input pricing rises from $0.20 to $0.40 per million tokens.
• Cache writes on GPT-5.6 Sol cost $5.00 for short context and $10.00 for long context.
• Regional processing endpoints are charged a 10% uplift for models released on or after March 5, 2026.
• GPT-5.6 Sol promotional pricing is available at least through November 21, 2026.
• Priority processing was renamed Fast mode on July 30, 2026.

This is the line most pricing round-ups get wrong, because they quote the short-context rate and stop there. A workload that routinely fills a large context window pays the long-context rate, not the headline one. At that point, the comparison with Claude Opus 5 inverts. Sol’s long-context input rate sits above the single price Anthropic lists for Opus 5, rather than below it as the short-context rate does.

How Many People Use ChatGPT vs Claude

• OpenAI’s country-level release covers more than 1 billion people putting ChatGPT to work.
• The dataset reflects messages sent within ChatGPT Free, Go, Plus, and Pro accounts, generally managed by individuals rather than organizations.
• Multimedia accounts for 7.8% of messages globally, the fastest-growing use case.
• More than one in ten messages are multimedia in countries including Brazil and Colombia.
• Individually managed ChatGPT Free, Go, Plus and Pro accounts now account for a 5% higher share of messages than they did 12 months earlier.
• In France and Czechia, the share of messages from users aged 35 or older increased by more than 10 percentage points in the past year.
• At work, people are more than twice as likely to use ChatGPT to complete a task or create something.
• Anthropic’s privacy-preserving telemetry continuously samples a slice of conversations every day. No user count appears in the Anthropic sources captured here.

Source: OpenAI Signals Q2 2026 release and Anthropic Economic Index, June 2026

No source captured for this article publishes a Claude user total. Circulating estimates have been left out here rather than repeated. Anyone wanting the fuller picture on each side can compare the ChatGPT numbers against Claude’s own disclosure record.

Benchmark Claims, Vendor by Vendor

• OpenAI reports GPT-5.6 Sol scoring 53.6 on Agents’ Last Exam, an evaluation of long-running professional workflows across 55 fields.
• Even at medium reasoning, OpenAI reports GPT-5.6 Sol beating Fable 5 by 11.4 points at roughly one-quarter the estimated cost.
• OpenAI reports GPT-5.6 Sol with max reasoning at 80 on the Artificial Analysis Coding Agent Index, 2.8 points above Fable 5.
• OpenAI reports BrowseComp at 92.2% and OSWorld 2.0 at 62.6% for GPT-5.6 Sol.
• Anthropic reports Claude Opus 4.6 outperforming GPT-5.2 by around 144 Elo points on GDPval-AA.
• Anthropic reports Opus 4.6 achieving the highest score on Terminal-Bench 2.0.
• Anthropic reports Claude Opus 5 scoring three times as high as the next-best model on ARC-AGI 3.
• Anthropic reports Opus 5 passing at around 1.5x the next-best model’s rate on Zapier AutomationBench for the same cost per task.

Source: OpenAI GPT-5.6 launch post and Anthropic Claude Opus 4.6 and Opus 5 launch posts, captured August 2026

Worth noting: Every row above is a vendor-reported result, run on a harness that vendor selected or on a third-party index the vendor scored itself against. OpenAI benchmarks GPT-5.6 Sol against Claude Fable 5, Anthropic benchmarks Opus 4.6 against GPT-5.2 on GDPval-AA, and the ARC-AGI 3 and Zapier AutomationBench rows name no comparator beyond next-best. Matched conditions never applied, so the two sets cannot be ranked against each other.

Is Claude more accurate than ChatGPT?

Neither company publishes a shared accuracy measure, so the answer depends entirely on the task and the harness. Anthropic claims leads on agentic coding and knowledge work, and OpenAI claims leads on long-horizon workflows and browsing. The narrower coding question is covered in more depth in this comparison of Claude and ChatGPT on coding benchmarks.

What People Actually Do With Claude

• Anthropic’s classifier identified 93% of Claude conversations as producing an artifact.
• The share of chat and Cowork conversations categorized as personal use spikes from around 35% on weekdays to just under 50% on weekends.
• Recipe requests are 2.3 times more frequent at 6 p.m. compared to the average.
• On April 14, tax-related clusters were eight times as common as on the average day in May.
• People ask for news at 7 a.m. local time.
• On weekends, the Claude Code usage clusters that fall the most include backend architecture, API debugging, and data storage.
• The clusters that increase the most on weekends include AI agent design, quant trading and gaming.
• Claude sessions now increasingly consist of long-running agentic tasks.

The weekday-to-weekend swing is the most useful line in the dataset. It shows how thin the boundary between a coding tool and a general assistant has become. That shift toward long-running delegation also appears across the broader agentic AI deployment data.

Where Claude Usage Runs Heaviest

• Australia records a Usage Index of 6.40 against an expected value of 1.
• Singapore ranks 2, and Switzerland ranks 3 on the same usage rank order.
• United States sits at 3.87, ranked 12 in the same order.
• United Kingdom is ranked 16.
• The most common topics in Australia start with Homework at 9.3%.
• A reading above 1 means a state uses Claude more than expected based on its population, and below 1 means less.
• Peru, Uruguay and Costa Rica rose the most among countries in global rankings.
• Usage in parts of Latin America, Oceania, and Africa increased faster than in other parts of the world.

Singapore ranks second on the same dashboard, but the captured view lists no index reading for it, so the table above covers only the countries carrying a published value. Both companies now publish geographic adoption data, and both point the same way: the early-adopter lead is narrowing. Denominators differ, though. Anthropic normalizes to population, while OpenAI reports country-by-country share of messages and global rankings, so the two leaderboards are not interchangeable.

Claude vs ChatGPT Pros and Cons

• Anthropic’s documented coding market share is an estimated 54%.
• OpenAI documents reach more than 1 billion people putting ChatGPT to work.
• OpenAI’s cheapest tier undercuts Anthropic’s at $0.20 per million input tokens.
• Anthropic’s cheapest tier costs $1 per million input tokens.
• Claude Fable 5 still falls back to Opus 5 for requests considered dual-use, including virology, toxicology, and molecular design.
• Claude Opus 5 remains behind Mythos 5 on cybersecurity tasks.
• OpenAI reports GPT-5.6 Sol with max reasoning coming within one point of Fable 5 on the Artificial Analysis Intelligence Index while completing tasks in 61% less time at roughly half the estimated cost.
• Claude Opus 5 at max effort performs within 0.5% of Fable 5’s peak score on CursorBench 3.2 at half the cost per task.

Source: Menlo Ventures 2025 enterprise survey, OpenAI Signals Q2 2026, and both vendors’ model documentation, August 2026

Read across the table and the pattern in the Claude vs ChatGPT statistics is consistent: Anthropic wins on measured enterprise spend and on coding, OpenAI wins on reach and on price at the bottom of the range. Neither pattern predicts which assistant produces better output on a given task, because no shared evaluation exists to test that.

AI Watermarking Under the EU AI Act

• Future Claude models will generate text containing a watermark to comply with the EU AI Act.
• As of August 2, the EU requires AI providers serving its market to mark AI-generated content.
• Other major model developers have signed the same Code of Practice and will implement their own watermarks.
• Nothing is added to the text, and there are no hidden characters.
• Watermarking requires no extra tokens and will not be more expensive.
• The watermark carries no identifying information and cannot be traced to a specific person, organization, or chat.
• Google DeepMind found no statistically significant differences from the unwatermarked model when testing SynthID-Text on Gemini traffic.
• That test served a watermarked model to a portion of Gemini traffic and compared thumbs-up and thumbs-down ratings against the unwatermarked model.

This is the first genuinely new axis of difference between the two assistants in a year, and it is a compliance axis rather than a capability one. Anyone building on either API inside the European Union now inherits a provenance obligation that did not exist in 2025.

Safety Fallbacks Still Downgrade Claude on Dual-Use Topics

• Anthropic reduced biology-related fallbacks by about 85% across its product surfaces.
• A fallback switches the system to a less capable model after a biology-related query.
• Anthropic intentionally launched Fable 5 with almost all biology queries blocked.
• That decision resulted in a high number of false positives for legitimate biology users.
• Fable 5 is not yet usable for professional biology research and drug development.
• Users should see far fewer fallbacks on lab-result interpretation, symptom questions, and educational biology.

Is Claude safer than ChatGPT?

Anthropic publishes more about its refusal machinery than OpenAI does, which makes Claude easier to audit rather than demonstrably safer. The roughly 85% fallback reduction is a measured change to a documented mechanism, and no equivalent public figure exists on the OpenAI side for comparison.

Which Is Better for Me, ChatGPT or Claude, in 2026?

The published data supports a split answer rather than a winner. For agentic coding or long-horizon engineering inside an enterprise, the spend evidence points to Claude, which holds an estimated 54% of the coding market, compared to 21% for OpenAI. For high-volume, cost-sensitive, or consumer-facing work, OpenAI’s floor of $0.20 per million input tokens and its broader consumer footprint carry more weight.

Context economics, rather than raw capability, usually decides everything in between. A workload that fills a large window pays GPT-5.6 Sol’s long-context input rate of $8.00 per million tokens, while Anthropic’s model overview lists one input price of $5 for Claude Opus 5. Published evaluations separate the flagships by a few points. Costs on a large-context workload separate them by a multiple.

Why Are People Moving From ChatGPT to Claude in the Enterprise?

Enterprise buyers moved on coding performance rather than general capability. Anthropic has held 18 months atop the LLM leaderboards for coding, and its coding share is up from 42% just six months ago, driven in large part by the popularity of Claude Code.

The shift is narrower than the headline implies. Google increased its enterprise share from 7% in 2023 to 21% in 2025, and the three companies together account for 88% of enterprise LLM API usage. Broader adoption patterns across the category sit in the wider AI coding adoption data, which shows the same concentration.

Conclusion

Anthropic’s estimated 40% share of enterprise LLM spend against OpenAI’s 27% remains the single most reliable number in this comparison, and it is also the narrowest. It measures API dollars from one survey, not users, not quality, and not consumer reach. On reach, OpenAI’s disclosure of more than 1 billion people remains the only user-reach figure anywhere in the captured source set. The honest read on the Claude vs ChatGPT statistics available today: two companies winning different markets, publishing different metrics to prove it.

What changes next is regulatory rather than technical. Watermarking obligations took effect for AI providers serving the EU market as of August 2; both vendors have signed the same Code of Practice, and provenance is becoming a shipped product feature rather than a research topic. Movement is most likely to appear next in the pricing gap at the low end and in context-window economics at the high end.