Which AI models do companies build on?
22,156 companies are hiring for LLM work right now, but few say which model the new hire will build on. We read every job posting that names one, threw out the noise, and counted 2,570 companies.
OpenAI 1,802 companies, Claude 1,576, Gemini 536. In January it wasn't close: Claude was named in 37% of builder postings then, and in 61% in October.
478 of OpenAI's 1,802 companies name it only as Azure OpenAI: about a quarter of its footprint arrives through Microsoft. A company that builds on two models counts for both. Cohere (19) and Grok (15) are too small to show.
| Group | Companies |
|---|---|
| OpenAI only | 806 |
| OpenAI and Claude | 996 |
| Claude only | 580 |
| Neither | 188 |
996 companies have open roles building on both; 806 on OpenAI but not Claude, 580 on Claude but not OpenAI, 188 on neither (Gemini or open-weight only).
Claude went from 37% to 61% of AI-builder postings this year; OpenAI from 83% to 63%
Share of job postings that build software on a model and name each provider, by month published, Jan–Oct 2026. Estimated from a labeled random sample of 2,655 builder postings; shaded bands are the 95% range. The open postings agree: among those published in May, Claude had 75% as many builder companies as OpenAI; among those from October, 94%.
| Month | OpenAI | Claude | Gemini | Open-weight |
|---|---|---|---|---|
| 2026-01 | 83% | 37% | 25% | 17% |
| 2026-02 | 82% | 47% | 26% | 18% |
| 2026-03 | 79% | 48% | 18% | 14% |
| 2026-04 | 78% | 63% | 18% | 9.8% |
| 2026-05 | 78% | 63% | 16% | 9.6% |
| 2026-06 | 72% | 61% | 17% | 6.2% |
| 2026-07 | 73% | 60% | 20% | 13% |
| 2026-08 | 62% | 57% | 25% | 15% |
| 2026-09 | 69% | 60% | 19% | 10% |
| 2026-10 | 63% | 61% | 17% | 4.8% |
Four in five companies hiring for AI-assisted coding ask for Claude Code
A different market from the one above: 5,635 companies hire people to write code with an AI assistant, and their software may not use AI at all. Cursor 2,765, GitHub Copilot 2,566, OpenAI Codex 1,492.
| Tool | Companies |
|---|---|
| Claude Code | 4522 |
| Cursor | 2765 |
| GitHub Copilot | 2566 |
| OpenAI Codex | 1492 |
| Gemini CLI | 134 |
| Windsurf | 133 |
Smaller companies lean to Claude, the largest to OpenAI
Share of model builders in each size band that build mostly on OpenAI or on Claude. The rest hire for several models with no clear favourite (45% of the smallest companies, 30% of the largest) or build mostly on Gemini or open-weight models.
| Company size | Companies | Mostly OpenAI | Mostly Claude | Mostly Gemini | No clear favourite |
|---|---|---|---|---|---|
| 1–200 | 1011 | 19% | 27% | 3.2% | 45% |
| 201–1,000 | 628 | 25% | 25% | 3.3% | 42% |
| 1,001–10,000 | 719 | 41% | 16% | 6.1% | 34% |
| 10,000+ | 208 | 42% | 7.7% | 10% | 30% |
Who builds on what
The 400 largest companies with a confirmed role, one tile each, sized by employee count and coloured by the model it builds on most. This is where OpenAI is strongest: 179 of them build mostly on OpenAI, 35 on Claude, 39 on Gemini, 22 on open-weight models, and 125 have no clear favourite (split tiles show their two main models). Paler = fewer open roles: a pale tile is one posting, not a company-wide bet.
| Company | Sector | Country | Employees | Builds on |
|---|---|---|---|---|
| Shell | Energy & utilities | United Kingdom | 220,144 | OpenAI, Claude, Open-weight, Cohere / Grok |
| HCA Healthcare | Health & life sciences | United States | 155,839 | Gemini |
| Orange | Telecom | France | 149,231 | Gemini, Open-weight |
| SAP | Software & internet | Germany | 147,392 | OpenAI |
| Honeywell Technologies | Manufacturing & industrial | United States | 137,612 | OpenAI, Gemini |
| Santander | Financial services | Spain | 131,300 | OpenAI, Claude, Gemini, Open-weight |
| Itaú Unibanco | Financial services | Brazil | 112,823 | OpenAI, Claude, Gemini, Open-weight |
| General Motors | Manufacturing & industrial | United States | 110,245 | Gemini |
| Ericsson | Telecom | Sweden | 106,853 | OpenAI |
| Roche | Health & life sciences | Switzerland | 106,716 | OpenAI |
| TD | Financial services | Canada | 105,534 | OpenAI |
| Morgan Stanley | Financial services | United States | 103,323 | OpenAI, Claude, Gemini, Open-weight |
| RBC | Financial services | Canada | 100,598 | OpenAI, Claude |
| Procter & Gamble | Manufacturing & industrial | United States | 97,648 | Gemini |
| Bayer | Manufacturing & industrial | Germany | 96,110 | OpenAI, Claude, Gemini |
| EDF | Energy & utilities | France | 94,358 | Open-weight |
| UnitedHealth Group | Health & life sciences | United States | 92,195 | OpenAI |
| Tencent | Software & internet | China | 89,561 | OpenAI, Gemini |
| Scotiabank | Financial services | Canada | 88,230 | OpenAI, Claude, Gemini, Open-weight |
| U.S. Bank | Financial services | United States | 87,453 | OpenAI |
| Barclays | Financial services | United Kingdom | 84,768 | OpenAI |
| A.P. Moller - Maersk | Transport & logistics | Denmark | 79,855 | OpenAI, Claude, Gemini, Open-weight |
| United Airlines | Transport & logistics | United States | 76,563 | OpenAI, Claude |
| Philips | Health & life sciences | Netherlands | 72,926 | OpenAI, Claude |
| Gap Inc. | Retail & consumer | United States | 72,621 | Gemini |
| BT Group | Telecom | United Kingdom | 69,170 | OpenAI, Claude, Open-weight |
| Deutsche Telekom | Telecom | Germany | 67,772 | OpenAI, Open-weight |
| Thales | Manufacturing & industrial | France | 65,936 | OpenAI, Claude |
| American Airlines | Transport & logistics | United States | 61,231 | OpenAI, Claude |
| BMO | Financial services | Canada | 60,986 | OpenAI |
| Johnson Controls | Manufacturing & industrial | Ireland | 60,723 | OpenAI, Claude, Open-weight |
| GE HealthCare | Health & life sciences | United States | 60,219 | OpenAI, Open-weight |
| Danaher | Health & life sciences | United States | 59,355 | OpenAI, Claude |
| Qualcomm | Software & internet | United States | 58,202 | OpenAI, Open-weight |
| Eli Lilly and Company | Health & life sciences | United States | 57,722 | OpenAI, Claude |
| Inspire | Retail & consumer | United States | 56,963 | OpenAI |
| Allstate | Financial services | United States | 56,952 | OpenAI |
| Qatar Airways | Transport & logistics | Qatar | 56,685 | Gemini |
| Intesa Sanpaolo | Financial services | Italy | 56,341 | OpenAI, Gemini |
| Renault Group | Manufacturing & industrial | France | 56,088 | OpenAI, Open-weight |
| BNY | Financial services | United States | 55,211 | OpenAI, Claude, Cohere / Grok |
| Boston Scientific | Health & life sciences | United States | 55,152 | OpenAI |
| IndusInd Bank | Financial services | India | 55,123 | OpenAI |
| Gallagher | Financial services | United States | 53,334 | OpenAI |
| AMD | Software & internet | United States | 51,744 | Open-weight |
| Atlas Copco Group | Manufacturing & industrial | Sweden | 50,452 | OpenAI |
| Ecolab | Manufacturing & industrial | United States | 50,262 | OpenAI |
| Humana | Financial services | United States | 49,247 | Gemini |
| Zurich Insurance | Financial services | Switzerland | 46,612 | Claude |
| Whirlpool Corporation | Manufacturing & industrial | United States | 46,014 | Gemini |
| Groupe BPCE | Financial services | France | 43,774 | Open-weight |
| The World Bank Group | Other | United States | 43,767 | OpenAI |
| Adobe | Software & internet | United States | 43,517 | OpenAI, Claude, Gemini, Open-weight |
| Royal Caribbean Group | Real estate & travel | United States | 43,347 | OpenAI |
| Magna International | Manufacturing & industrial | Canada | 42,504 | OpenAI |
| Henkel | Manufacturing & industrial | Germany | 42,503 | OpenAI, Open-weight |
| Amgen | Health & life sciences | United States | 40,850 | OpenAI, Claude, Gemini, Open-weight |
| JD.COM | Software & internet | China | 39,584 | OpenAI, Claude |
| Mapfre | Financial services | Spain | 39,498 | OpenAI |
| Johnson & Johnson Innovative Medicine | Health & life sciences | 39,046 | Claude |
OpenAI leads in 12 of 13 sectors, Claude in the UK and Germany
Gap between the share of AI-building companies that build on Claude and on OpenAI, in percentage points. The widest gap is in manufacturing & industrial: 38 points in OpenAI's favour.
| Segment | Companies | Claude | OpenAI |
|---|---|---|---|
| All model builders | 2570 | 61% | 70% |
| Software & internet | 1121 | 69% | 68% |
| Financial services | 447 | 58% | 71% |
| Manufacturing & industrial | 210 | 40% | 78% |
| Health & life sciences | 195 | 61% | 73% |
| Retail & consumer | 105 | 61% | 62% |
| Education & research | 89 | 60% | 71% |
| Business & legal services | 58 | 57% | 83% |
| Real estate & travel | 56 | 62% | 70% |
| Media, gaming & marketing | 51 | 59% | 75% |
| Transport & logistics | 50 | 64% | 80% |
| Telecom | 43 | 47% | 72% |
| Energy & utilities | 42 | 60% | 64% |
| Government & nonprofit | 39 | 46% | 67% |
| United States | 1328 | 66% | 69% |
| United Kingdom | 122 | 67% | 63% |
| Germany | 121 | 65% | 62% |
| India | 104 | 52% | 84% |
| Canada | 84 | 60% | 75% |
| France | 81 | 59% | 62% |
| Brazil | 58 | 53% | 60% |
| Singapore | 49 | 49% | 78% |
Only 1 in 8 postings that name ChatGPT or OpenAI is about building software on it
Every posting that names a provider, split by what the role actually does with it. Only the dark segment counts as building software on the model. Claude is named in more postings than OpenAI (26,684 against 22,256), mostly because of Claude Code: 33% of its mentions are a coding assistant.
| Provider | Postings naming it | Builds software on it | Codes with it | Office use | Listed only | Resells / implements | AI-trainer gig | Other mention |
|---|---|---|---|---|---|---|---|---|
| Claude (Anthropic) | 26684 | 9.2% | 33% | 33% | 18% | 2.0% | 0.1% | 4.9% |
| OpenAI | 22256 | 13% | 11% | 35% | 22% | 4.6% | 0.1% | 15% |
| Gemini (Google) | 8065 | 9.4% | 2.7% | 28% | 29% | 3.4% | 0.0% | 27% |
| Llama (Meta) | 1185 | 23% | 0% | 0.1% | 27% | 8.2% | 0% | 42% |
| Mistral | 427 | 36% | 0% | 0.5% | 25% | 10% | 0% | 29% |
| DeepSeek | 297 | 39% | 0% | 9.4% | 38% | 4.0% | 1.3% | 7.7% |
| Qwen | 239 | 51% | 0% | 0.8% | 43% | 4.6% | 0.8% | 0% |
| Grok (xAI) | 227 | 7.9% | 0.9% | 12% | 21% | 19% | 1.3% | 37% |
| Cohere | 181 | 12% | 0% | 0.6% | 10% | 15% | 0% | 62% |
Numbers you can take with you
Copy any of these with a link back. Please cite “Echoloc Live” and the date: the numbers move every week.
2,570 companies are hiring people to build software on a named AI model (Echoloc Live, AI Model Wars, Nov 8, 2026).
OpenAI 1,802 companies, Claude 1,576. 478 of the OpenAI companies name it only as Azure OpenAI: about a quarter of OpenAI's footprint arrives through Microsoft.
Four in five companies hiring for AI-assisted coding ask for Claude Code: 4,522 of 5,635 (Echoloc Live, Nov 8, 2026).
Claude was named in 37% of AI-builder job postings in January 2026 and in 61% in October.
Company size decides the bet: among AI builders with 10,000+ employees, 42% build mostly on OpenAI and 7.7% mostly on Claude; under 200 employees it is 19% and 27%.
Only 1 in 8 job postings that mention ChatGPT or OpenAI is about building software on it. The rest are office use, coding assistants or a line in a skills list.
Download, cite, talk to us
Download and cite
Weekly aggregates under CC BY 4.0: counts by model, sector, size and country, the monthly trend, coding assistants and the noise breakdown. Company lists are not in the file.
Download CSV data as of Nov 8, 2026Charts are free to use with a credit: every chart has a link, and the map has PNG and embed buttons.
Talk to us
We read every message. Tell us what looks wrong, what you would measure next, or ask for a cut of the data.
- Feedback or a correctionSomething looks off, or a company is missing or misread
- A custom cutA country, a sector, or your own list of companies
- Press and researchMethodology questions, quotes, embargoed numbers
- Suggest the next dashboardWhat should we track the same way?
hello@echoloc.ai
Get the companies behind the numbers
Every company here has a dated, public job posting. In echoloc you can filter them by AI stack, sector, size and country, and export the list.
How this is measured
This dashboard measures one thing: which AI models companies are hiring people to build software on, right now. Not who signed a contract, not which chatbot employees open. A company counts for a provider when it has at least one open job posting whose duties include building, integrating, fine-tuning, deploying or operating that provider's models inside the company's products or internal tools. Writing code with an AI assistant is measured separately (see “Coding assistants”): the software a developer writes with Claude Code need not use Claude at all.
Source and period
Public job postings collected daily by echoloc. Every number is about postings open on Nov 8, 2026, the date of the latest data: 32% of them were published in the previous 30 days, 45% 31 to 90 days earlier and the rest before that. It is a snapshot of what companies are hiring for now, refreshed every week, not a count of new postings that week. The trend chart is the exception: it covers postings published each month since January 2026.
From 4 million postings to 2,570 companies
- 3,981,067 open postings at 367,428 companies.
- 63,266 of them name an AI model or coding assistant: OpenAI (ChatGPT, GPT, o-series, Azure OpenAI, Codex), Anthropic (Claude, Claude Code, and the model names Opus, Sonnet and Fable), Google (Gemini, Gemma), Meta Llama, Mistral, Qwen, DeepSeek, Cohere, xAI Grok, GitHub Copilot, Cursor, Windsurf.
- Consultancies, IT services, staffing, outsourcing and agencies are removed (18,829 postings: they build for clients or hire on their behalf, so the stack is the client's choice). They are caught by industry and, where the industry field misses them (an IT-services firm filed under “Software Development”, a recruiting marketplace under “Internet”), by an LLM reading the company's own description, plus a hand check of the largest employers on the map; as are the providers themselves and their affiliates (1,713 postings: OpenAI, Anthropic, Google, Microsoft, Meta, Amazon, Mistral, Cohere, xAI and SpaceX, and others), AI-trainer gigs and placeholder employers such as “Stealth Startup”.
- An LLM reads every remaining posting and decides, provider by provider, what the role does with it (below).
- Result: 2,570 companies building software on a named model, 5,635 hiring people to code with an AI assistant.
For scale: 22,156 companies have an open posting about LLM or generative-AI work, and most never name the model. This dashboard can only see the ones that do.
Classification
Naming a model is not building on it. “Comfortable with ChatGPT” is office software; “Claude” is also a first name; “Gemini” is a zodiac sign; “no ChatGPT during the interview” is a rule. Every posting that names a provider was read by an LLM (Claude Haiku 5.5) against a fixed rubric, using the excerpts around each mention, and labeled per provider as:
- builds software on it: integrates, fine-tunes, deploys or operates the provider's models in the company's products or internal tools (“integrate the OpenAI API into our claims workflow”, “build agents on Claude”);
- codes with it: uses the provider's coding assistant to write software (Claude Code, Codex, Gemini CLI);
- office use: uses or rolls out the chat assistant for people's own work (“use ChatGPT to draft documents”, “administer Claude Enterprise”);
- listed only: named in a generic list of LLMs, with no sign the role builds with it (“familiarity with LLMs such as GPT, Claude, Llama”);
- AI-trainer gig, resells / implements the provider for clients, or other mention (company boilerplate, interview rules, a person named Claude).
Two passes, two levels of evidence
Every “builds software on it” verdict goes through a second, stricter pass with one narrow question per provider: does the text show this role integrating the provider's models into software? This week it re-checked 9,899 first-pass verdicts. 62% were confirmed: the posting describes building on the provider. Where the strict pass found only a list (“hands-on with OpenAI and Anthropic APIs, LangChain, vector databases”), the verdict counts as likely if the role itself is a builder role (engineer, developer, architect, data scientist, AI/ML) and is dropped otherwise. Sales, marketing, recruiting, consulting and business-management titles never qualify. Of the 2,570 companies, 2,134 have at least one confirmed role; the others count on likely roles only. Admin roll-outs, coding-assistant lists and other mentions always move to their own “not counted” bucket.
Provider groups and the Azure question
OpenAI includes ChatGPT, GPT and o-series models, Codex and Azure OpenAI, because Azure OpenAI serves OpenAI's models. That channel matters: 478 companies mention OpenAI only as “Azure OpenAI”. Cloud platforms are treated alike: a posting that names only AWS Bedrock or only Google's Vertex AI could mean Claude, Gemini, Llama or a dozen other models, so it is not attributed to anyone (Microsoft's Azure OpenAI is different: it serves only OpenAI's models). Google means Gemini and Gemma. Open-weight groups Meta Llama, Mistral, Qwen and DeepSeek. On the map and in the size chart, a company counts for the model it builds on most, and as “no clear favourite” when no model has at least twice the roles of the next.
Counting
Every number is companies, not postings, unless it says otherwise, and one open role is enough to count in the totals: 80% of the companies have exactly one. A role is a distinct job title: the same opening posted in several cities counts once. One posting proves a company is hiring for it, not that the whole company has bet on a model, so the map shows evidence strength: it includes only companies with at least one confirmed role, and tiles get paler the fewer roles stand behind them (1, 2–3, 4+); every company stays in the charts and in the map's search. Shares by sector count a company once per provider it builds on, so rows can add up to more than 100%. Size bands and tile sizes use the number of employees on the company's public profile. That count can include freelancers, drivers or hosts who list the company (a marketplace can show 180,000 people against an official 501–1,000 staff), so it is capped at the top of the company's official size range, and a company with no count takes the bottom of its range. Sectors group self-reported industries into thirteen buckets.
Trend
Labeling every posting since January would cost far more than it would teach, so the monthly trend comes from a deterministic random sample of up to 2,500 postings per month that name a provider, labeled with the same two passes. The lines are the share of builder postings naming each provider group, with a 95% confidence band. Index volume differs by month, so only shares are compared, never raw counts.
Limitations
- Hiring is a leading signal of building, not a measure of spend or tokens. A company with a mature, fully staffed AI team may hire less than one starting out.
- Postings that describe AI work without naming a model (“experience with LLM APIs”) are invisible here.
- LLM labels are not perfect. We read samples by hand before every release; the remaining errors are mostly borderline “builds on” vs “listed only”, which is why the two evidence levels are reported.
Questions or a custom cut (your sector, your country, your competitors): hello@echoloc.ai.
Questions people ask
Which AI model do companies build on the most?
As of Nov 8, 2026, OpenAI: 1,802 companies are hiring people to build software on it, then Claude (1,576) and Gemini (536). 478 of the OpenAI companies mention it only as Azure OpenAI.
Is Claude or OpenAI more popular with enterprises?
OpenAI, by a wide margin at the top: among AI builders with 10,000+ employees, 42% build mostly on OpenAI and 7.7% mostly on Claude. Among companies under 200 employees Claude is slightly ahead.
Is Claude catching up with OpenAI?
Yes. In a labeled sample of builder job postings, Claude's share rose from 37% in Jan to 61% in Oct, while OpenAI's moved from 83% to 63%.
What does “build software on a model” mean here?
The job posting says the role integrates the model into the company's software: calls its API, builds an agent, RAG system or AI feature on it, fine-tunes or hosts it. Using ChatGPT at work, or writing code with Claude Code, is counted separately.
Which AI coding assistant do companies ask for in job postings?
Claude Code (4,522 companies), then Cursor (2,765) and GitHub Copilot (2,566).
Why not just count mentions of ChatGPT in job postings?
Because most of them are not about building. Only 13% of postings naming OpenAI or ChatGPT describe building software on it; the rest are office use, coding assistants, skills lists or rules like “no ChatGPT during the interview”.
How often is this updated, and what period does it cover?
Weekly. Each release covers job postings that are open on the date of the data (Nov 8, 2026 for this one), most of them published in the previous three months.
Can I use these numbers or charts?
Yes, under CC BY 4.0: cite “Echoloc Live, AI Model Wars” with the date and a link. The embed button gives an iframe that credits the source automatically.