OpenAI vs Anthropic
Compare commercial API data controls, buyer checks and dated public adoption evidence. Inspect what the sources establish before choosing a provider.
OpenAI vs Anthropic for a commercial API shortlist
Direct commercial API evaluation; consumer subscriptions and third-party routes have separate terms. Next editorial review due . This edition records what was checked on the review date; it does not promise current contract status. If the review is overdue, use the linked primary sources to recheck before acting.
| Decision | OpenAI | Anthropic | Buyer check |
|---|---|---|---|
| Model training | API inputs and outputs are not used for model training by default; explicit opt-in sharing is an exception.Official OpenAI documentation ↗ | Commercial inputs and outputs are not used for training by default; feedback or other permission can allow training use.Official Anthropic documentation ↗ | Check the purchased product, sharing settings and feedback controls. |
| Retention and feature eligibility | Abuse-monitoring logs are retained up to 30 days by default, with stated exceptions; application-state retention depends on the feature. Approved retention controls also have endpoint and model limitations.Official OpenAI documentation ↗ | Zero data retention requires enablement and applies only to eligible features. Some models, stateful features, flagged content and legal holds have separate retention rules.Official Anthropic documentation ↗ | List the exact model, endpoint, tools and storage features; obtain the applicable retention terms before evaluating sensitive data. |
How to turn this into a purchase decision
Run both providers on the same approved task set and record task success, latency, failure handling and total cost, including tools and storage. Check deployment route, data location, support and contract terms separately. This page has not run a comparative model benchmark or verified account-specific prices.
Evidence appendix: three rechecked disclosures
Intercom, Linear and Sentry each named both providers in the reviewed scope on October 15. This supports co-disclosure only. It does not establish the model, request route, spend, relative quality or joint production usage. Publication and adoption dates remain unknown unless explicitly recorded; Intercom’s source has a June 25, 2026 effective date.
| Company | Vendor / signal | Industry / scope | Observed (UTC) | Evidence |
|---|---|---|---|---|
| Intercomintercom.com | OpenAIPresence observed | Customer service softwareDefault US-hosted service: AI processing | First observed: 2026-09-10 | Subprocessor disclosureEvidence notesThe published subprocessor list names this provider for the stated processing purpose. Source publication: Not stated. Source effective date: 2026-06-25. Confirmed adoption date: Not established. |
| Intercomintercom.com | AnthropicPresence observed | Customer service softwareDefault US-hosted service: AI processing | First observed: 2026-09-10 | Subprocessor disclosureEvidence notesThe published subprocessor list names this provider for the stated processing purpose. Source publication: Not stated. Source effective date: 2026-06-25. Confirmed adoption date: Not established. |
| Linearlinear.app | OpenAIPresence observed | Developer toolsAI processing | First observed: 2026-09-10 | Subprocessor disclosureEvidence notesThe published subprocessor list names this provider for the stated processing purpose. Source publication: Not stated. Confirmed adoption date: Not established. |
| Linearlinear.app | AnthropicPresence observed | Developer toolsAI processing | First observed: 2026-09-10 | Subprocessor disclosureEvidence notesThe published subprocessor list names this provider for the stated processing purpose. Source publication: Not stated. Confirmed adoption date: Not established. |
| Sentrysentry.io | OpenAIPresence observed | Developer toolsAI processing | First observed: 2026-09-10 | Subprocessor disclosureEvidence notesThe published subprocessor list names this provider for the stated processing purpose. Source publication: Not stated. Confirmed adoption date: Not established. |
| Sentrysentry.io | AnthropicPresence observed | Developer toolsAI processing | First observed: 2026-09-10 | Subprocessor disclosureEvidence notesThe published subprocessor list names this provider for the stated processing purpose. Source publication: Not stated. Confirmed adoption date: Not established. |
No confirmed OpenAI-to-Anthropic or Anthropic-to-OpenAI migration is established by this appendix.
Real public-source records, manually reviewed. This is a dated snapshot, not a continuous live feed or a representative market survey.
Counts describe the current snapshot. The time window applies to movement history below.
Distinct companies naming OpenAI in covered disclosures.
Distinct companies naming Anthropic in covered disclosures.
3 of 3 companies naming at least one provider. Co-disclosure is a potential co-adoption signal, not confirmed joint production usage.
Industry mix in this sample
Changes in the last 12 months
History unavailable. This window runs from 2025-10-15 to 2026-10-15. Our reviewed baseline begins 2026-09-10. Additions, removals, switches, and net change cannot yet be calculated. This does not mean there were zero changes.
Requires matched, dated before-and-after sources.
Co-disclosure alone does not show a switch.
Initial observations are not new adoptions.
Companies and source disclosures
These lists also serve as the OpenAI and Anthropic customer evidence for this comparison.
| Company | Vendor / signal | Industry / scope | Observed (UTC) | Evidence |
|---|---|---|---|---|
| Intercomintercom.com | OpenAIPresence observed | Customer service softwareDefault US-hosted service: AI processing | First observed: 2026-09-10 | Subprocessor disclosureEvidence notesThe published subprocessor list names this provider for the stated processing purpose. Source publication: Not stated. Source effective date: 2026-06-25. Confirmed adoption date: Not established. |
| Intercomintercom.com | AnthropicPresence observed | Customer service softwareDefault US-hosted service: AI processing | First observed: 2026-09-10 | Subprocessor disclosureEvidence notesThe published subprocessor list names this provider for the stated processing purpose. Source publication: Not stated. Source effective date: 2026-06-25. Confirmed adoption date: Not established. |
| Linearlinear.app | OpenAIPresence observed | Developer toolsAI processing | First observed: 2026-09-10 | Subprocessor disclosureEvidence notesThe published subprocessor list names this provider for the stated processing purpose. Source publication: Not stated. Confirmed adoption date: Not established. |
| Linearlinear.app | AnthropicPresence observed | Developer toolsAI processing | First observed: 2026-09-10 | Subprocessor disclosureEvidence notesThe published subprocessor list names this provider for the stated processing purpose. Source publication: Not stated. Confirmed adoption date: Not established. |
| Sentrysentry.io | OpenAIPresence observed | Developer toolsAI processing | First observed: 2026-09-10 | Subprocessor disclosureEvidence notesThe published subprocessor list names this provider for the stated processing purpose. Source publication: Not stated. Confirmed adoption date: Not established. |
| Sentrysentry.io | AnthropicPresence observed | Developer toolsAI processing | First observed: 2026-09-10 | Subprocessor disclosureEvidence notesThe published subprocessor list names this provider for the stated processing purpose. Source publication: Not stated. Confirmed adoption date: Not established. |
Direct vs. cloud-mediated usage
3 companies name a provider directly in their disclosure. The reviewed records do not establish whether requests go directly to the provider or through Azure OpenAI, Amazon Bedrock, or Vertex AI. No delivery-route percentages are inferred. An AWS infrastructure disclosure alone is not evidence of a particular Bedrock model.
OpenAI and Anthropic: company-by-company evidence matrix
Inspect the actual source, scope, delivery channel and observation date behind each provider relationship.
Define the comparison before counting
This frozen export contains 6 provider observations for 3 distinct companies, selected from the site’s merged evidence batches through October 15, 2026. It keeps only provider IDs openai and anthropic. Source dates in the downloads are preserved; publishing this guide is not a new review of every source. A missing provider record means not observed in this export, not that the company does not use it.
The company matrix
Each cell is a dated observation. Use the downloadable record-level CSV for source URLs and the delivery field; Azure OpenAI or another hosted route must not be silently relabeled a direct provider contract.
| Company | OpenAI evidence | Anthropic evidence |
|---|---|---|
| Intercom | Default US-hosted service: AI processing (review 2026-10-15) | Default US-hosted service: AI processing (review 2026-10-15) |
| Linear | AI processing (review 2026-10-15) | AI processing (review 2026-10-15) |
| Sentry | AI processing (review 2026-10-15) | AI processing (review 2026-10-15) |
Turn a row into a defensible answer
For a buyer asking “Does this company use both?”, answer with the named company, disclosure scope and observation date. Then open the source to check today’s wording. A subprocessor entry supports a processing relationship in the stated scope; it does not establish spend, model version, exclusive use or company-wide adoption.
- Deduplicate by company domain before counting companies.
- Report both-provider overlap separately; adding the two counts double-counts shared companies.
- Require dated, same-scope before/after evidence before claiming a provider was replaced.
- Use task-specific trials for product quality. This matrix is not a model benchmark.
Sources and reproducibility
Continue your task
Anthropic vs OpenAI: what are we comparing?
The unit of comparison is the named provider in a company’s public subprocessor list. Each company counts once per provider, even if the list repeats it across regions. A company appearing on both sides contributes once to the combined denominator.
This initial sample is deliberately small and selected for inspectable evidence. Its co-disclosure rate describes only the included companies. It is not an enterprise AI market-share estimate, a revenue comparison, or a measure of model usage.
Claude vs OpenAI: company, model, or product?
Claude is Anthropic’s AI product and model family. OpenAI is a provider whose products include ChatGPT. Comparing Claude with OpenAI can mix a product decision with a provider decision. This page keeps the observation at the provider level.
For a product-versus-provider distinction, see Anthropic vs ChatGPT. For product selection, evaluate the actual task, available integrations, data controls, and commercial terms. A disclosure does not identify every model version or establish product performance. The linked source records support the adoption comparison; they do not supply a model benchmark.
What would count as an observed switch?
A credible switch claim needs explicit evidence that one provider replaced another for the same service scope. Matched disclosure removals and additions alone show a document change and are not sufficient to confirm a switch. A provider being named for the first time in our collection is only a first observation. Even a documented disclosure change does not establish spend, cancellation, or replacement across the entire company.
Explore AWS customer evidence for the infrastructure layer, or read our research on interpreting vendor changes.
What the evidence can tell you
A public job board supports an ATS observation. A subprocessor disclosure identifies a provider that may process data for particular services. A vendor customer story describes a published relationship. These sources have different scopes and are labeled separately.
Counts cover the records on these pages, not the whole market. No observation in our covered sources does not mean a company does not use a vendor. Source recency and unresolved conflicts affect confidence; observations do not establish spend, renewal timing, or customer churn.
Need evidence for your shortlist?
Tell us the company, vendor and service scope you need. Request a coverage review or discuss product access.