LLM Market Share 2026: Enterprise Usage & Web Traffic

Direct Answer: Which LLM Has the Largest Market Share?

There is no single LLM market-share leader across every definition. Menlo Ventures estimated Anthropic at 32% of enterprise LLM usage in mid-2025, ahead of OpenAI at 25% and Google at 20%. In consumer web traffic, ChatGPT remained much larger: it represented 67.07% of December visits across seven standalone AI domains in Similarweb’s public table, while Gemini represented 21.09%.

Those percentages measure different populations. Menlo models enterprise usage and spend using a survey of more than 150 technical leaders. Similarweb estimates global browser visits to named domains; it excludes apps, APIs, embedded assistants and many in-product experiences. Adding or averaging the two systems would produce a meaningless number.

This maintained index publishes 34 source-backed rows with a definition attached to every share. Data period: mid-year and full-year 2025. Last verified: July 11, 2026.

Cite This Index

Canonical URL: https://konabayev.com/blog/llm-market-share/

Recommended citation: Tugelbay Konabayev, “LLM Market Share 2026: Enterprise Usage & Web Traffic,” Konabayev.com, July 11, 2026, https://konabayev.com/blog/llm-market-share/

Machine-readable versions:

• CSV dataset
• JSON dataset
• JSONL claims feed

For one figure, cite the stable claim fragment. For example, #llmshare-003 is Anthropic’s enterprise usage estimate and #llmshare-024 is ChatGPT’s derived share of the seven-domain web set.

LLM Market Share Summary

Measurement system	Leader	Share	Population
LLMSHARE-003 enterprise usage	Anthropic	32%	Menlo enterprise usage estimate
LLMSHARE-004 enterprise usage	OpenAI	25%	Menlo enterprise usage estimate
LLMSHARE-005 enterprise usage	Google	20%	Menlo enterprise usage estimate
LLMSHARE-024 standalone web	ChatGPT	67.07%	Seven-domain December visit set
LLMSHARE-025 standalone web	Gemini	21.09%	Seven-domain December visit set

The enterprise rows are Menlo’s modeled percentages. The web rows are Konabayev calculations using the seven December visit counts that Similarweb published. They are a share of that defined set, not Similarweb’s estimate of every AI interaction.

Enterprise LLM Usage Share

Enterprise usage is a modeled deployment measure, not consumer popularity or audited provider revenue. Preserve the survey and market-model context when citing these shares.

Menlo reported that OpenAI’s early enterprise lead declined from 50% at the end of 2023 to 25% in mid-2025. Anthropic moved to 32%, while Google reached 20%. Meta Llama and DeepSeek remained smaller in the same estimate.

Claim	Provider/model family	Enterprise usage share
LLMSHARE-006	Meta Llama	9%
LLMSHARE-007	DeepSeek	1%

The published values do not sum to 100% because the public highlights do not list every remaining provider or category. Do not renormalize the visible rows unless the analysis explicitly defines the omitted group.

Menlo is an investor in Anthropic and other AI companies. That conflict does not make the data unusable, but it increases the importance of preserving the method and comparing future editions rather than treating the result as audited vendor revenue.

Enterprise API Spend and Workload Mix

Menlo estimated model API spending at $8.4 billion, more than double its $3.5 billion estimate six months earlier. The research combined survey responses with market modeling; the value is not a sum of public provider financial statements.

Claim	Enterprise market finding
LLMSHARE-001	Survey population: 150+ technical leaders.
LLMSHARE-002	Modeled API spend: $8.4B, up from $3.5B.
LLMSHARE-008	Anthropic code-generation usage: 42%, versus 21% for OpenAI.
LLMSHARE-009	Open-source workload share: 13%, down from 19%.
LLMSHARE-015	Startup builders with majority-inference workloads: 74%, up from 48%.
LLMSHARE-016	Large enterprises with most compute in inference: 49%, up from 29%.

Usage share, code-generation share, workload architecture and API spend remain separate metrics. A provider can lead one use case without leading every enterprise deployment. An open-source workload can also run through a cloud endpoint, making vendor and model-family attribution non-trivial.

Model Switching Benchmarks

Provider switching was much less common than model-version upgrading inside the same provider. This separates provider retention from rapid model churn.

Most surveyed builders changed models within their existing provider rather than switching vendors. This suggests that provider-level retention can coexist with rapid model-version churn.

Claim	Switching behavior	Share
LLMSHARE-010	Upgraded inside the existing provider	66%
LLMSHARE-011	Did not switch models	23%
LLMSHARE-012	Switched vendors	11%

Menlo also reported a fast migration inside Anthropic after Claude 4 launched:

Claim	Model-version result
LLMSHARE-013	Sonnet 4 reached 45% of Anthropic users within one month.
LLMSHARE-014	Sonnet 3.5 declined from 83% to 16%.

This is one provider and one launch window, not a universal half-life for LLMs. It does demonstrate why a static model-ranking article can become obsolete quickly while a provider relationship remains stable.

Consumer Web Traffic by Standalone AI Domain

The web dataset measures estimated visits to seven named domains and omits app, API and embedded-product usage. It is useful only when that denominator remains explicit.

Similarweb published monthly estimates for seven domains. Their December values total 8.227 billion estimated visits. ChatGPT contributed 5.518 billion and Gemini 1.735 billion.

Claim	Domain	December 2025 estimated visits
LLMSHARE-017	chatgpt.com	5,517,989,010
LLMSHARE-018	gemini.google.com	1,735,320,959
LLMSHARE-019	deepseek.com	328,865,716
LLMSHARE-020	grok.com	271,154,744
LLMSHARE-021	perplexity.ai	179,583,644
LLMSHARE-022	claude.ai	172,690,841
LLMSHARE-023	meta.ai	21,559,214

The visits do not equal users. One person can generate many visits, and usage inside mobile apps, operating systems, social feeds, developer tools, search results and APIs may never reach the standalone domain.

Derived December Web Share

The derived percentages are reproducible shares of the seven published December rows. They are not a claim about every consumer or enterprise LLM interaction.

The following table divides each December visit count by the sum of the same seven rows. It is reproducible from the public source but deliberately narrow.

Claim	Platform	Share of seven-domain December visits
LLMSHARE-026	DeepSeek	4.00%
LLMSHARE-027	Grok	3.30%
LLMSHARE-028	Perplexity	2.18%
LLMSHARE-029	Claude	2.10%
LLMSHARE-030	Meta AI	0.26%

The market became less concentrated in this defined web set during 2025. In January, ChatGPT represented roughly 84% of the same seven-domain total; by December, the derived share was 67.07%. That does not mean its visits fell year over year. Similarweb reported 43% January-to-December growth. Faster growth by competitors changed the mix.

2025 Web-Traffic Growth

Growth rate shows momentum, while share shows relative size inside a declared market set. A platform can lead one measure and not the other.

Claim	Platform	January-to-December growth reported by Similarweb
LLMSHARE-031	Gemini	548%
LLMSHARE-032	Claude	125%
LLMSHARE-033	Perplexity	80%
LLMSHARE-034	ChatGPT	43%

Growth percentage and market share answer different questions. A small platform can grow quickly while remaining small, and the largest platform can add more absolute visits with a lower growth rate.

How to Define LLM Market Share

A valid market-share statement names the metric, numerator, denominator, period and excluded usage. Without those fields, the percentage cannot be reproduced or compared.

Choose one numerator and denominator before publishing a ranking:

Definition	Numerator	Denominator	Key blind spot
Enterprise usage	Workloads or usage attributed to a provider	Surveyed or modeled enterprise usage	Sampling and model routing
API spend	Estimated spend attributed to model APIs	Total modeled API spend	Private pricing and cloud resale
Web traffic	Estimated visits to a named domain	Visits to a declared domain set	Apps, APIs and embedded usage
Developer adoption	Developers reporting use	Survey respondents	Self-report and survey frame
Revenue	Provider/model revenue	Defined market revenue	Bundled products and private companies

Never present one denominator as the entire LLM market without naming what is excluded. For a recurring index, preserve historical snapshots rather than silently replacing old values, and record source-version changes separately from market movement.

For adjacent evidence, see AI search statistics, AI search referral traffic benchmarks, AI code assistant statistics, and the LLM SEO guide.

Why Enterprise and Consumer Leaders Differ

Enterprise selection rewards integration, performance, security, contracts and workload economics, while standalone web traffic rewards consumer distribution and habit. The same provider can therefore occupy very different positions across the two systems.

An enterprise may access a model through a cloud platform, internal gateway or application without visiting the model provider’s website. It may route tasks across several models and record the gateway vendor rather than the underlying model. Contract pricing, data residency, support, throughput and governance can matter as much as a public chatbot experience.

Consumer web traffic behaves differently. A product with a well-known standalone destination can accumulate billions of visits, while an assistant embedded in an operating system, search engine, social network or productivity suite may create significant usage that the standalone-domain table does not see. Mobile apps introduce another missing surface.

This is why ChatGPT can lead the defined web set while Anthropic leads Menlo’s enterprise estimate. The two statements do not conflict. They describe different buyers, access paths and units. A defensible article should present both and resist the urge to announce one universal winner.

Reproduce the Seven-Domain Share Calculation

The December web-share table can be reproduced directly from Similarweb’s seven visit rows. Sum the published December visits, then divide each domain’s value by that total.

The denominator is 8,227,164,128 estimated visits. ChatGPT’s 5,517,989,010 divided by that total produces 67.07% after rounding to two decimals. Gemini’s 1,735,320,959 produces 21.09%. The same calculation is used for the five smaller domains.

Do not add a new platform to the numerator without also adding it to the denominator and documenting the source. Do not compare a future eight-domain share with this seven-domain history as though the composition stayed constant. If Similarweb changes a domain, methodology or geographic scope, record a series break.

The downloadable CSV preserves both source visit counts and derived share rows. The evidence field marks the derived calculations, while source-published values remain identifiable. This distinction lets another analyst reproduce the arithmetic without attributing Konabayev’s calculation to Similarweb.

Use the Index for Market Decisions

Choose the measurement system that matches the decision rather than the most dramatic percentage. Product strategy, developer relations, media planning and infrastructure procurement require different views.

For enterprise model procurement, track workload-level quality, latency, price, reliability, data handling, deployment path and switching cost. Menlo’s provider shares are market context, not a substitute for a task-specific evaluation. A 32% share does not establish that one provider is best for your code, language, risk or latency requirement.

For consumer distribution, web traffic can identify large destinations and fast-growing competitors. Add app usage, referral behavior, engagement, geography and audience overlap before translating traffic into a marketing plan. Visits alone do not reveal retention, paid conversion or query intent.

For investors or market sizing, neither dataset is sufficient by itself. Enterprise modeled usage is not audited revenue, and web traffic is not spend. Use provider financial disclosures, cloud-marketplace data and consistent revenue definitions where available, while labeling estimates and bundled products.

For content and SEO, the traffic mix shows where audiences conduct visible standalone sessions. The Similarweb source table can support platform-specific discovery hypotheses, but referral traffic and citation behavior require separate measurement.

Maintenance Rules for a Fast-Moving Index

The page should update only when a comparable source row exists, not whenever a provider announces a launch. Product news can explain movement, but it is not market-share evidence.

Web rows can update quarterly if Similarweb publishes a consistent table. Preserve the prior monthly observations and calculation denominator. Enterprise rows should update when Menlo Ventures publishes a comparable survey or market model, with sample and method changes clearly marked.

Do not silently replace model families, domain definitions or market buckets. If an assistant moves from one domain to another, note the continuity problem. If a provider bundles several models, keep provider and model-family shares distinct. If a new source measures tokens or API calls, create a new measurement block rather than placing it beside spend or visits as though the units match.

A stable yearless canonical preserves citations. Dates belong on the observations and update log, not in a new URL every quarter. This makes historical claims auditable while allowing the index to remain current.

Methodology and Limitations

Every row is labeled as enterprise usage, API spend, workload mix, web visits, derived web share or growth. Cross-definition percentages are never added or averaged.

The enterprise section uses Menlo Ventures’ public mid-year 2025 report. The web section uses Similarweb’s published monthly table. The seven-domain shares were calculated as domain December visits / sum of seven December visit rows, rounded to two decimals.

Menlo combines a survey with proprietary market modeling and has portfolio interests in AI companies. Similarweb estimates web traffic rather than reporting first-party product analytics. Neither source independently measures the full LLM market. No enterprise and consumer percentages are combined.

This page will retain one yearless canonical. Web-traffic rows should be reviewed quarterly when a comparable official table is available; enterprise rows should update when Menlo publishes a methodologically comparable edition.

FAQ

What is the most popular LLM in 2026?

It depends on the population. ChatGPT led the seven-domain consumer web set, while Anthropic led Menlo’s mid-2025 enterprise usage estimate.

What is ChatGPT’s market share?

ChatGPT represented 67.07% of December 2025 visits across the seven standalone domains in Similarweb’s table. This is not a share of apps, APIs or all AI usage.

What is Anthropic’s enterprise LLM market share?

Menlo estimated Anthropic at 32% of enterprise usage in mid-2025, ahead of OpenAI at 25% and Google at 20%.

Are open-source LLMs gaining enterprise share?

Not in the cited Menlo snapshot. It estimated 13% of workloads used open-source models, down from 19% six months earlier.

Can web visits measure LLM revenue?

No. Web visits exclude paid API consumption, embedded products and contract value. Revenue and traffic require different datasets.

How often should an LLM market-share index update?

Review web traffic quarterly and enterprise usage when comparable source editions appear. Keep historical data because models and traffic can shift within weeks.

Sources

• Menlo Ventures, 2025 Mid-Year LLM Market Update
• Similarweb, Winners and Losers in the Gen AI Market