hopala AI implementation agency עב Get in touch

A capability review and comparison

This page doesn't crown a winner; it tells you when to reach for which, and every row is backed by a primary source.

As of

Which one, when

A real Monday-morning task, and the tool you open for it.

  • A 4,000-row spreadsheet to analyse

    Both run Python on the file in a sandbox. ChatGPT swallows bigger spreadsheets; Claude works even on a free account.

    Both

  • You need an image for a post

    ChatGPT. Claude doesn't generate images — it builds charts and diagrams in code, not illustrations.

    ChatGPT

  • Deep research with sources, before a meeting

    ChatGPT lets you approve the research plan up front, restrict it to named domains and export to Word and PDF; Claude's docs describe none of those steps.

    ChatGPT

  • Wire the bot into your internal systems

    Both speak MCP. Claude accepts any remote server by pasting a URL; ChatGPT publishes apps through a review portal.

    Both

  • A long PDF with tables and charts to understand

    Claude — it reads the visuals inside the PDF on any plan; on ChatGPT that's Enterprise-only.

    Claude

  • Show it what's on your screen while you talk

    ChatGPT. Its voice mode can see live camera or a shared screen; Claude's voice only listens.

    ChatGPT

  • A live editable panel beside the chat, on a phone too

    Claude — Artifacts run on the mobile apps; ChatGPT's Canvas is still desktop-only.

    Claude

  • Code to write and review

    Both ship a real coding agent. Codex is included even on the free plan; Claude Code starts at Pro.

    Both

Product & app capabilities

The stuff you actually touch, day to day.

Reading images and charts inside uploaded documents

Most teams upload reports full of charts — the question is whether the AI actually sees them or just pulls the text.

Claude

Reads text and the visuals — images, charts, graphics — in PDFs under 100 pages.1Above 100 pages the docs promise text only; no plan gate appears anywhere on the page.

No plan gate stated in the docs

ChatGPT

Visual Retrieval — reading visuals inside a PDF — is Enterprise-only and PDF-only; on every other plan it's text only, images discarded.2Even on Enterprise, PDFs added as GPT knowledge or Project files are text-only.

Enterprise only; excluded on Free, Pro, Team, Edu

Claude — For visuals inside a PDF, Claude documents the capability with no plan gate up to 100 pages, while on ChatGPT it's Enterprise-only and text everywhere else.

Deep / agentic research mode

Both run multi-step research; the difference is control over the run and the format you can share the result in.

Claude

Agentic research that chains searches and returns answers with checkable citations; needs web search turned on.3Can also draw on connected Gmail, Google Calendar and Google Docs. No plan-review step, domain allow-list or report export documented.

Pro, Max, Team, Enterprise

ChatGPT

Proposes a research plan to approve up front, can be interrupted and narrowed to sources, and exports to Markdown, Word and PDF.4You can restrict the run to domains you enter, or just prioritise them. Uses read-only actions from connected apps.

Varies by plan and country

ChatGPT — ChatGPT documents four steering controls Claude's docs don't — editing the plan, interrupting mid-run, restricting to named domains, and export to Word and PDF.

Image generation

Any real image work — a post, a banner, a mockup — has to go elsewhere if the tool can't draw.

Claude

None. Claude doesn't generate images; it builds diagrams, charts and interactive visuals in HTML and SVG.5A positively stated absence, not an inference.

ChatGPT

ChatGPT Images 2.0 on all tiers: region editing, conversational editing, an aspect-ratio picker, transparent backgrounds and in-image text.6Images with thinking on Plus, Pro and Business. Every generated image lands in an Images library.

Base on all tiers; with thinking on Plus/Pro/Business

ChatGPT — Claude doesn't generate images at all, and says so.

Browser control / agentic web use

Both can drive a browser; choosing a Chrome extension vs a built-in browser changes your security surface.

Claude

Claude in Chrome — an extension that reads, clicks and navigates alongside you: workflow recording, multi-tab, console/network/DOM reads, scheduled tasks and 1Password login (beta, macOS).7Chrome only. Anthropic says the extension is still risky and requests sensitive permissions like debugger and nativeMessaging.

All paid plans; allow/block-lists on Team and Enterprise

ChatGPT

Control runs through the desktop app and Codex rather than a standalone browser: a built-in browser, Record & Replay, scheduled tasks and admin Lockdown Mode.8The dedicated Atlas browser retires on 9 Aug 2026. Some capabilities are region-limited (EU, UK, Switzerland).

Mixed: scheduled tasks on Plus and up; Record & Replay excludes EU/UK/Switzerland

Claude — A narrow Claude win on shape, not substance: one shipped Chrome extension on every paid plan, versus ChatGPT's real but scattered capabilities across the desktop app and Codex, plus a dedicated browser that's retiring.

Code-execution sandbox for uploaded data

Both run Python on your files; the size caps and the sandbox's network access change both capability and leak risk.

Claude

Yes — code execution and file creation on every plan; up to 30MB per file into the sandbox; network egress is an admin setting.9On Free/Pro/Max egress is on by default; Team is package-managers-only; Enterprise is off. Anthropic recommends starting with egress off.

All plans, including Free

ChatGPT

Yes — Advanced Data Analysis runs Python in a stateful Jupyter environment; ~50MB per spreadsheet (512MB for other types); the sandbox has no internet at all.10No admin toggle — the environment is simply offline. Up to 80 files per 3 hours, 25GB per user.

Plus, Pro, Enterprise; Free limited to 3 uploads a day

Depends — It depends on priorities: ChatGPT takes bigger spreadsheets and its Python is unconditionally offline, while Claude ships the sandbox on every plan and hands admins the egress decision.

Video and spoken-audio generation

If you need multimedia output — video or speech — a tool that makes neither forces a second tool.

Claude

No — Claude generates neither video nor speech.5Its visual output is code-based: diagrams and charts as HTML/SVG. No video, image or speech endpoint in the API.

Not applicable — absent on every plan

ChatGPT

Yes to both — Sora 2 video and native speech output, though the video API is being retired.11sora-2 and sora-2-pro make 16- and 20-second clips; the Videos API shuts down 24 Sep 2026. Speech via gpt-audio-1.5 and Realtime.

Not verified — OpenAI's pricing pages returned 403; figures from docs only

ChatGPT — One vendor ships video and speech and the other ships neither — but the Videos API's September 2026 shutdown makes the video half of that lead unstable.

Voice conversation in the app

Both let you talk to the assistant; only one can also see your screen or camera while you talk.

Claude

Voice mode (beta) — spoken conversation only: you talk, Claude replies aloud, can run a web search mid-chat, and you switch freely between text and voice.12No live camera video or screen-share inside a voice conversation, at any tier.

All plans including Free; beta

ChatGPT

Voice in the apps, with one thing Claude's lacks: Advanced Voice Mode shares live camera or your screen mid-conversation so ChatGPT sees what you see.13Shipped first to Free, Plus, Pro and Team on iOS and Android; stricter regions (EU, Switzerland) followed later.

Live video/screen in voice — mostly paid; staged by region

ChatGPT — On in-app voice specifically: ChatGPT's voice can see live camera or your screen while you talk, whereas Claude's voice only listens.

Connector / app ecosystem — how it reaches your systems

Claude accepts any remote MCP server openly; ChatGPT routes apps through a review portal — a reach-vs-curation trade-off.

Claude

Open MCP: any remote MCP server can be added as a custom connector, plus a browsable connectors directory.14You paste a server URL (optionally your own OAuth); on Team and Enterprise only Owners add org connectors.

Free (1 connector), Pro, Max, Team, Enterprise

ChatGPT

Apps/plugins in ChatGPT, also built on MCP via the Apps SDK, but published through an OpenAI submit-and-review portal.15Distribution is curated: submit, approve, publish; strong apps may earn directory placement or proactive suggestions.

Not documented — availability and admin controls unverified from a primary source

Depends — Both sit on MCP, so the split is governance not protocol: Claude's open-URL connectors reach an internal API fast with no gatekeeper, while ChatGPT's review model gives you a vetted catalogue.

Free tier: what an unlicensed employee actually gets

Sets what staff can do before you pay and — critically for an employer — whether their chats train the model by default.

Claude

$0. Near-full surface, small tank: web search, file creation and code execution, memory, extended thinking and connectors including MCP. The frontier model (Fable 5) is gated.16A session-based limit that resets every five hours. Chats train the model only if the user turns it on in Privacy Settings.

Free; no Fable 5, Projects, Research, Claude Code

ChatGPT

$0. GPT-5.5 Instant (limited) and GPT-5 Thinking Mini; 27K context, ~12 pages of input. Limited on nearly everything; includes web search, Projects and using GPTs.17Content trains OpenAI's models with opt-out. Ads may appear on the free tier in some countries.

Free; building GPTs and scheduled tasks start at Plus

Depends — ChatGPT's free tier hands out more surface (Projects, GPTs, images), while Claude's gives full connector/MCP reach and — decisively for an employer — doesn't train on chats unless the employee opts in, the reverse of ChatGPT's opt-out default.

Projects/Spaces: a container for chats, knowledge and instructions

A mid-size team gets a reusable, context-preserving workspace, so onboarding to a body of work doesn't mean restating the background each chat.

Claude

Yes — Projects are workspaces with their own chat history, a knowledge base and per-project instructions.18The knowledge base is shared across every chat in the project; each project has its own memory. Enhanced retrieval on paid plans.

All plans (Free capped at 5 projects)

ChatGPT

Yes — Projects group chats, reference files and per-project instructions, with built-in project memory.19You can also paste Google Drive files and Slack channel links as sources. Unlimited number of projects.

All plans, free and paid

Tie — Both ship a real project container — grouped chats, knowledge base, instructions and memory — on every tier including free. A tie on the core concept.

A reusable custom assistant a non-developer can build and share — and where

Decides whether non-technical staff can stand up reusable assistants themselves, and whether they can be shared only inside the company or to the public.

Claude

Yes, no code — two paths: Projects (instructions + a knowledge base, point-and-click) and Skills (a skill.md file, optionally with files). Sharing is org-internal only; no public store.20Projects work on Free; Skills on every plan, with org-wide distribution via an admin flow.

Projects and Skills on all plans; org distribution on Team/Enterprise

ChatGPT

Yes, no code, via the GPT Builder (a Custom GPT = instructions + knowledge + actions). Building requires a paid plan; you can publish to the public GPT Store.21Free users can discover and use GPTs but not build them. Public publishing needs a verified Builder Profile.

Building: paid (Plus/Team/Enterprise); using: Free

Depends — It depends on the goal: for public distribution or the smoothest build UX, ChatGPT and the GPT Store; for a free, org-internal reusable assistant, Claude builds one on the free tier.

Generative side-panel: a live editable document/code beside the chat (Artifacts vs Canvas)

A side-by-side editable panel lets teams iterate on drafts and code in place — but only one of them keeps working on a phone.

Claude

Artifacts: a dedicated pane beside chat for code, docs, sites and interactive apps; live-editable with a version selector. Works on the mobile apps too.22Sharing splits by tier: public Publish on Free/Pro/Max; org-only Share on Team/Enterprise.

All plans, including mobile

ChatGPT

Canvas: a side workspace to co-write docs and code, with highlight-to-edit and writing/coding shortcuts; exports to PDF, Markdown and Word. Desktop and web only.23OpenAI's own Canvas page lists Web, Windows and macOS; mobile is still coming soon — it doesn't run on the iOS/Android apps.

All plans, but desktop/web only

Depends — Both ship a genuine live-editable side-panel, at parity on web and desktop; the difference is platform — Claude's Artifacts work on mobile, while ChatGPT's Canvas is desktop-only.

Max file size per file (consumer/Team apps)

The per-file ceiling decides whether a big report, dataset or scanned document goes in as one upload or has to be split.

Claude

500MB per file in chat; 30MB per file for Project knowledge files.1The chat ceiling is high, but inside a Project it drops to 30MB.

ChatGPT

512MB per file (hard limit); images 20MB; spreadsheets ~50MB; text files also capped at 2M tokens each.10A dense document can hit the token cap well before 512MB.

Tie — Raw ceilings are essentially equal (500MB vs 512MB); each side has a different secondary catch — Claude drops to 30MB in Projects, ChatGPT adds a 2M-token cap that can bind first.

Scheduled / recurring tasks that run on their own

A mid-size company can offload recurring work — morning digests, KPI reports, competitor monitoring — without a human triggering it each time.

Claude

Yes — Scheduled tasks in Claude Cowork. Describe a task once; it runs on a cadence (hourly, daily, weekly) or on demand.24Runs remotely, so it fires even when your computer is asleep. Cowork is in beta, rolling out from the Max plan.

All paid plans; rolling out from Max

ChatGPT

Yes — Scheduled Tasks in ChatGPT. Schedule one-off and recurring prompts (reminders, digests, monitoring); managed at chatgpt.com/schedules.25Runs server-side with no app open; a task can't run more than once an hour.

Go, Plus, Pro, Business, Enterprise (not free)

Tie — Both run autonomous tasks server-side on paid tiers; capability is at parity, and the difference is maturity — ChatGPT's is generally available, Claude's is newer and in beta.

Mobile apps (iOS/Android): availability, voice, and the modality gap vs desktop

A team picking a default phone assistant cares whether staff can talk to it, point a camera at a problem, or pull their calendar and inbox by voice.

Claude

Native iOS and Android apps. Voice mode (beta) on both, on every plan; voice can run web searches, and on paid tiers reach connected Gmail, Calendar and Drive.12No live camera video or screen-share inside voice, at any tier.

Voice free and paid; voice into Gmail/Calendar/Drive — paid

ChatGPT

Native iOS and Android apps. Standard voice for all free users; advanced speech-to-speech hears tone; paid tiers add live camera video and screen sharing in a voice chat.13Video/screen-share is capped per day on eligible plans.

Standard voice free; live video/screen — paid, daily-capped

Depends — The baseline is a tie — real-time voice on both; they diverge on extras: ChatGPT adds camera and screen in voice, Claude adds voice access to Gmail/Calendar/Drive. Pick by which matters more.

Shareable link to a single conversation (public vs org-only)

A mid-size company needs to circulate an AI answer internally without leaking it onto the open web.

Claude

Yes. Share creates a link to a snapshot of the chat. Free/Pro/Max can share publicly; Team and Enterprise only with org members.26The snapshot includes messages and Artifacts; attached files are excluded and stay private.

All tiers; public only on Free/Pro/Max

ChatGPT

Yes. Share creates a unique URL. Free/Plus links open for anyone; Team/Business and Enterprise are limited to workspace members.27On Enterprise you can't share outside the workspace; an outsider gets a no-access page.

Free/Plus public; Team/Business/Enterprise org-only

Tie — Both do the same thing at the same tier class: consumer tiers allow public links, and the business tiers lock shared links to the workspace.

SCIM user provisioning and directory sync (Okta/Entra)

A company on Okta or Entra can auto-create — and, critically, auto-revoke — access as staff join and leave; on both this requires Enterprise.

Claude

Yes — automated SCIM provisioning and deprovisioning; the IdP pushes adds, removes and edits. Okta and Entra ID are explicitly supported.28Gated to Enterprise and Console orgs only; not on Team. Setup runs through WorkOS.

Enterprise only (not Team)

ChatGPT

Yes — automated SCIM provisioning and deprovisioning; Okta and Entra ID (Azure AD) explicitly supported, alongside more IdPs.29Gated to Enterprise and Edu only; not Business/Team. Okta syncs immediately, most others every 30–40 minutes.

Enterprise and Edu only (not Business/Team)

Tie — Both offer full SCIM provisioning and deprovisioning, both support Okta and Entra, both set up through WorkOS, and both gate it to the top tier. Like for like.

Audit logs + compliance API (eDiscovery/DLP/SIEM)

Legal, security and compliance teams need a programmatic, exportable record of every interaction — table stakes for a regulated mid-size company.

Claude

Yes — a Compliance API on Claude Enterprise. An Activity Feed logs who/what/when; separate endpoints retrieve (and hard-delete) chat, file and project bodies. Feeds SIEM.30Activity records are retained for 6 years by Anthropic and queryable within about a minute.

Claude Enterprise

ChatGPT

Yes — a Compliance Platform on ChatGPT Enterprise and Edu. An immutable event log plus a stateful point-in-time API; wired into eDiscovery/DLP/SIEM.31Logs are kept only 30 days on the platform — you must export continuously for longer history.

ChatGPT Enterprise or Edu

Claude — Both ship a full Enterprise compliance API with SIEM/DLP export; the verifiable edge is retention — Claude keeps audit records for 6 years, while OpenAI's platform holds 30 days and expects you to build your own export.

Admin roles model (built-in + custom RBAC) at Enterprise

Determines how cleanly an org can delegate admin, billing and analytics-only access without over-granting.

Claude

Four standard roles on Team and Enterprise: User, Admin, Owner, Primary Owner. Enterprise adds custom roles at the group level.32No dedicated built-in analytics-only role — you build one as a custom role on Enterprise.

4 roles on Team+Enterprise; custom on Enterprise

ChatGPT

Four built-in roles on Business and Enterprise/Edu: Owner, Admin, Analytics Viewer, Member. Plus custom member RBAC.33Includes a built-in read-only Analytics Viewer role; custom RBAC is Enterprise/Edu/Teachers only.

4 roles on Business+Enterprise/Edu; RBAC on Enterprise/Edu

Tie — Both ship four built-in roles at Enterprise and add custom roles; ChatGPT has a built-in Analytics Viewer, while Claude reaches the equivalent only via a custom role. Functionally comparable.

Admin-configurable data/chat retention window (Enterprise)

A company under GDPR or data-minimization needs to set and prove a retention window without a sales cycle — both let an admin do it in-product.

Claude

Yes — self-serve. An Owner sets a custom window in Organization settings; documented 30-day minimum; default is indefinite until you set one.34Out-of-window data is permanently deleted on save; every change is written to the audit logs.

Enterprise; requires the Owner role

ChatGPT

Yes — self-serve. Workspace admins control retention for Enterprise (and Edu, Healthcare, Business); deleted data is removed within 30 days.35The pages checked confirm the control but don't state a minimum window or the config location.

Enterprise/Edu/Healthcare and Business

Tie — Both offer a self-serve admin retention window at Enterprise with a ~30-day purge; Claude's docs are more specific, but that's a documentation edge, not a capability gap.

SOC 2 Type II attestation

SOC 2 Type II is the baseline attestation most procurement and security teams require before approving a vendor.

Claude

Yes — SOC 2 Type I and Type II, for the commercial products (Claude for Work and the Anthropic API).36Consumer Free/Pro are out of scope. Report copies via the Trust Portal.

Commercial products only

ChatGPT

Yes — a SOC 2 Type 2 report (and SOC 3).37Covers the API Platform, ChatGPT Enterprise, Edu and Team; report period January–June 2025.

API Platform, Enterprise/Edu/Team

Tie — Both hold a current SOC 2 Type II report covering their business tiers; OpenAI additionally publishes a SOC 3.

Admin, security & privacy

What IT will ask before signing.

SAML single sign-on (SSO) — tier required

Both include SSO on business tiers via DNS domain verification; table stakes for IT, and effectively a tie.

Claude

Available on Team, Enterprise and Console orgs; needs a DNS TXT verification record; setup runs through WorkOS.38IdP guides: Okta, Entra ID, Google, OneLogin, JumpCloud, Duo. Enforcement is split across two switches — for Claude and for Console.

Team, Enterprise and Console

ChatGPT

Available on Business, Enterprise and Edu, for both ChatGPT and the API Platform; needs a DNS TXT record.39One org ID supports a single IdP config. Asymmetry: turning on SSO on the API strips password login from every user on the domain.

Business, Enterprise, Edu

Tie — Both ship SAML SSO on the business tier (Claude: Team; ChatGPT: Business) behind DNS verification, and both cover the API side; the only real asymmetry is operational.

Is business data used for training by default

Both say no on paid business tiers; Claude's is contractually explicit, while ChatGPT's ChatGPT-side page couldn't be independently verified.

Claude

No. Anthropic: by default it won't use inputs or outputs from its commercial products to train. The Commercial Terms make it contractual.40Symmetric carve-out: report feedback explicitly and it may be used. Owners can switch it off org-wide. The customer keeps rights to inputs and owns outputs.

Commercial products: Claude for Work, API, Claude Gov

ChatGPT

No for the API — verified verbatim: since March 2023, API data isn't used to train unless you opt in. The same stance is stated for the business tiers, but that page wouldn't load to verify.41Symmetric carve-out here too — training only on explicit opt-in. The ChatGPT-side stance is asserted but unverified here (403).

API (verified); business tiers — asserted, unverified here

Tie — Both are no-training-by-default on paid business tiers, and both keep the same opt-in carve-out; Anthropic's prohibition is written into its Commercial Terms, and OpenAI's parallel couldn't be checked because the page is blocked — so no edge either way.

IP allowlisting

Enterprise-only on both; ChatGPT's is self-serve while Claude's needs a call with a rep — a setup-friction difference for locked-down networks.

Claude

Enterprise only, IPv4 and IPv6 (CIDR), and not self-serve — enabled by contacting an Anthropic rep.42

Enterprise (not self-serve)

ChatGPT

Enterprise and Edu only, off by default, self-serve, effective in 5–15 minutes; separate lists for the workspace and the Compliance API.43

Enterprise/Edu

ChatGPT — Same Enterprise gating, but OpenAI's is admin-configurable while Anthropic's requires a call with a rep.

Buying / running via AWS, Google Cloud or Azure

Claude sells its frontier line on all three clouds; ChatGPT is deep only on Azure — decisive if you must draw down committed cloud spend.

Claude

On all three major clouds: AWS (Amazon Bedrock, plus Claude Platform via AWS Marketplace), Google Cloud (Vertex AI) and Microsoft Foundry (Azure billing).44AWS and Azure Marketplace routes meter to the cloud bill, so spend can hit an existing commitment. Some features (Batch, MCP) are missing on Bedrock and Vertex.

Developer/API and enterprise procurement; unrelated to consumer plans

ChatGPT

Deep on Azure (Foundry sells the frontier GPT line including GPT-5.6), partial on AWS Bedrock (GPT-5.5/5.4 and gpt-oss), and on Google Cloud only as open-weight gpt-oss — not a managed frontier API.45On Bedrock the frontier access trails Azure by a release; on Google Cloud OpenAI appears only as open weights.

Developer/API and enterprise procurement; unrelated to consumer plans

Claude — Claude's frontier models are purchasable on all three major clouds, including Microsoft's own; OpenAI's frontier is full on Azure, a release behind on Bedrock, and absent as a managed API on Google Cloud.

Admin control over connectors and agent write-actions

ChatGPT gives admins a workspace default and confirmation on write actions; Claude's per-tool control is a developer setting, not an admin console.

Claude

Owners gate who can add connectors; per-tool allow/deny is a developer setting, not an admin-console toggle.46On Team/Enterprise only Owners add connectors. MCP servers can write; Anthropic's advice for agentic runs is to manually disable write tools, not a switch. The API side does support per-tool allow/deny.

Owner-only connector installs on Team/Enterprise; per-tool allow/deny on the API

ChatGPT

Admins set a workspace-wide default plus per-app overrides, and write actions require in-conversation confirmation.47A tiered write-permission ladder; the default requires confirmation for consequential changes (send, delete, post, purchase). Members don't get an always-allow option.

Default and per-app overrides on Business and Enterprise

ChatGPT — OpenAI documents an admin-set default, a per-app write-permission ladder and a blanket confirmation before writes; Anthropic gates who installs a connector but leaves turning off the write tools as advice to the user, not a control the admin holds.

Uptime commitments and status history

Neither offers a public SLA; ChatGPT's published 90-day API uptime runs higher, relevant for reliability-sensitive rollouts.

Claude

No public SLA. The status page shows 99.5% 90-day API uptime. Priority Tier targets 99.5% — but is closed to new purchase.48The Commercial Terms carry no uptime commitment or credits. New guaranteed capacity needs a sales contract.

Priority Tier is API-only, closed to new purchase

ChatGPT

No public SLA either. The status page shows 99.93% API uptime for April–July 2026. Priority processing is self-serve pay-as-you-go, but capacity isn't committed.49Priority is switched on per request at a per-token premium; above 1M TPM a sharp spike may drop to standard.

Priority processing on paid API accounts, no contract

ChatGPT — Neither publishes an SLA, but OpenAI's priority capacity can still be bought self-serve and its published API uptime is higher, while Anthropic's Priority Tier has closed to new purchase and routes you to sales.

Coding-agent containment and security lockdown

Both sandbox local runs; what matters is what a security admin can lock (domain allow-lists) and that PHI is excluded from cloud modes.

Claude

Runs locally; an OS-enforced sandbox (Seatbelt on macOS, bubblewrap on Linux/WSL2) with default-deny network egress and an admin-lockable domain allow-list. ZDR on request; explicitly not covered by HIPAA readiness.50Admins enforce policy centrally (failIfUnavailable, allowManagedDomainsOnly). Anthropic admits the proxy doesn't terminate TLS, so a broad entry can open a leak path.

The sandbox is built into every install; ZDR needs sales approval

ChatGPT

Local mode runs in the same OS primitives (Seatbelt/bubblewrap), default workspace-write, network off; Codex cloud runs on OpenAI infra with an admin-set domain allow-list. The BAA covers Codex Local only — don't use Codex cloud with PHI.51Enterprise admins set sandbox mode, approval policy and network centrally; for Codex cloud you pick domain presets and can restrict to GET/HEAD/OPTIONS.

Managed config and the HIPAA path are admin/Enterprise features

Depends — On containment they're near-indistinguishable — both wrap local runs in Seatbelt and bubblewrap with deny-by-default egress and an admin-locked allow-list; the split is regulatory: OpenAI documents a BAA path for Codex Local, while Anthropic states plainly that Claude Code isn't covered under HIPAA readiness.

EU data residency and zero-data-retention (ZDR)

ChatGPT can run inference in the EU; Claude is US-only — often the single deciding fact for EU-data-bound organisations.

Claude

US-only inference, no EU option; ZDR exists but is sales-gated.52The inference_geo parameter accepts only us and global. ZDR excludes Batch, the Files API, code execution and managed agents. EU residency is reachable only indirectly via Bedrock or Google Cloud.

Data residency on any API org (1.1x price); ZDR — sales only

ChatGPT

EU inference and storage available (eu.api.openai.com); ZDR exists but is approval-gated.41Residency is also supported in the UK, Switzerland, UAE and more. Residency endpoints add 10% for models from 5 March 2026.

Residency and ZDR need prior OpenAI approval

ChatGPT — Both charge ~10% for residency and both gate zero-retention behind sales, but only OpenAI can currently run the inference itself inside the EEA — Anthropic's own docs cap inference_geo at us and global.

Availability in Israel, Hebrew UI, and ₪ billing

Both sell in Israel with no Hebrew UI; only ChatGPT lists ILS as a supported billing currency, easing local procurement and VAT.

Claude

Sold in Israel; no Hebrew UI; USD list price, local currency where supported but no published currency list.53Localization covers 11 languages — Hebrew isn't one. Invoices via Stripe; no Israel-specific VAT guidance.

Consumer and Team via Stripe; Enterprise is quoted

ChatGPT

Sold in Israel; no Hebrew UI; the shekel (₪) is an explicitly supported billing currency.54Localization covers ~57 languages — Hebrew isn't one. The currency locks at signup; changing it means cancel and re-subscribe.

ILS pricing for Go, Plus, Pro and Business bought on web

ChatGPT — Both are sold in Israel and neither localizes into Hebrew, so the only difference is that OpenAI publishes ILS as a supported billing currency and Anthropic publishes no currency list at all.

Plans & pricing

The numbers as the official pricing page prints them.

Team / Business per-seat price and seat limits

Identical $20/seat headline; Claude caps a Team at 150 seats while ChatGPT Business publishes no maximum, which matters as you scale.

Claude

Claude Team, standard seat: $20/seat/month billed annually, $25 monthly; minimum 2 members, maximum 150 seats.16A flat per-seat fee with a weekly usage cap; you can enable credits for overage. A Premium seat also exists at $100/$125.

Team

ChatGPT

ChatGPT Business, standard seat: $20/user/month billed annually, $25 monthly; minimum 2 seats, no published maximum.57A flat per-user fee with baseline limits; you can buy credits for overage. From 19 Aug 2026 extra seats are charged immediately and prorated.

Business

Tie — Identical price ($20 annual / $25 monthly) and an identical 2-seat minimum; the one difference is the ceiling — Claude Team caps at 150 seats while Business publishes no maximum, so a 200-person company can self-serve on OpenAI but must move to Claude Enterprise.

Individual paid tier price and billing cadence

Both $20/month; only Claude offers a discounted annual prepay, ChatGPT is monthly-only — a small budgeting difference at volume.

Claude

Pro: $20/month billed monthly, or $200 up front annually (shown as $17/month).16The same page reconciles the two figures: $17 with an annual subscription, $20 if monthly.

Pro

ChatGPT

Plus: $20/month, billed monthly. OpenAI states it offers no annual billing or prepay for Go, Plus or Pro.58That one sentence covers all three consumer tiers; ChatGPT Business does offer annual billing, but it's a business plan.

Plus

Claude — Identical $20 monthly headline; Anthropic adds an annual prepay and OpenAI's own help centre says it offers none on any consumer tier.

High-usage individual tier

Both offer $100 and $200 power tiers pitched as 5x/20x usage; effectively a tie for heavy solo users.

Claude

Max 5x at $100/month and Max 20x at $200/month, monthly-only, described as 5x and 20x more usage per session than Pro.59The pricing page shows the Max card as from $100; the $200 figure appears only in the help centre.

Max 5x / Max 20x

ChatGPT

ChatGPT Pro at $100/month and $200/month, monthly-only, described as 5x and 20x more usage than Plus.58The $200 plan stays the highest-usage tier; you can switch between the two anytime.

Pro $100 / Pro $200

Tie — Identical price points ($100 and $200), the same 5x/20x structure, both monthly-only, and both show only the entry price on the page — though the multipliers aren't in the same unit (Anthropic per session vs Pro, OpenAI usage vs Plus).

Is the coding agent included on the free plan

Codex is bundled free with any ChatGPT plan; Claude Code starts at Pro, so free trial-ability of the coding agent differs.

Claude

No — Claude Code starts at Pro. The pricing page lists it under Pro, Max, Team and Enterprise; the Free plan has no Claude Code entry.16Free users aren't shut out of coding help in chat (generate code, run code) — what they don't get is the terminal agent.

Pro, Max, Team, Enterprise — not Free

ChatGPT

Yes — Work mode and Codex are included in every ChatGPT plan (Free, Go, Plus, Pro, Business, Edu, Enterprise), with no separate Codex subscription.60OpenAI scopes the free tier to quick coding tasks, so free access isn't parity with the paid tiers.

All plans including Free; Free scoped to quick tasks

ChatGPT — On the literal question, Codex is documented as included on Free and Claude Code isn't — though OpenAI itself limits free Codex to quick coding tasks rather than the full agent.

Seat prices for tiers that include the coding agent

Maps which subscription unlocks Claude Code vs Codex, and at what per-seat cost across the ladder.

Claude

Free $0 (no Claude Code). Pro $17 annual/$20 monthly. Max from $100. Team (2–150) $20/seat annual or $25 monthly, Premium $100/$125. Enterprise $20/seat + usage at API rates. Claude Code on Pro, Max, Team and Enterprise.16The docs state plainly that Claude Code requires Pro, Max, Team, Enterprise or Console — not Free.

Claude Code gated to Pro/Max/Team/Enterprise; excluded from Free

ChatGPT

Free $0; Go $8; Plus $20; Pro from $100; Business $20/user annual (2+) or $25 monthly; Enterprise/Edu contact sales. Codex is included from Free through Enterprise.60Free and Go carry no published limits; they're described only as quick/lightweight coding tasks.

Codex on every tier including Free; limits scale by plan

Depends — As a business seat the two are identical to the dollar — $20 annual, $25 monthly for 2+ — so an org buying seats has no price gap; OpenAI is cheaper only below the business ladder, where its $0 and $8 rungs still include Codex.

Published usage limits for the coding agent

ChatGPT publishes numeric per-window limits; Claude uses relative language plus dollar-planning figures, so buyers can't directly compare a seat's allowance.

Claude

No absolute numbers on any surface. The pricing page is relative only: more usage, 5x/20x over Pro, and so on; every tier carries a usage-limits-apply footnote.61What is published are dollar-planning figures — about $13 per developer per active day and $150–250 per month, 90% under $30 a day.

Every tier described relatively; Enterprise moves to seat price + usage at API rates

ChatGPT

Numeric ranges published, scoped to Codex and Work mode. Tables give local messages, cloud chats and code reviews per 5-hour window, per model.60The ranges are wide — the top of a band is roughly 6x the bottom — and the weekly limits on top are never quantified. More transparent than nothing, but not a cap you can budget against.

Plus and Business share one set; Pro splits into 5x/20x

ChatGPT — Even like for like — Codex docs against Claude Code docs — OpenAI publishes numeric per-model ranges and Anthropic publishes none, though ranges with unquantified weekly caps are a modest win.

Models & coding agents

Who runs what, and for how much.

Open-weight models you can self-host

OpenAI ships Apache-2.0 gpt-oss weights you can run yourself; Claude offers none — relevant for air-gapped or cost-controlled self-hosting.

Claude

None. Anthropic publishes no open-weight models; every Claude model is served through the API or a cloud marketplace.62Anthropic has released small open-weight tools for interpretability research, but no runnable Claude model.

ChatGPT

gpt-oss-120b and gpt-oss-20b, Apache 2.0 licensed; the 120b fits on a single H100 GPU. No per-token price — you run the weights.63

ChatGPT — OpenAI publishes usable open weights and Anthropic doesn't.

Maximum output length per response

Both cap around 128k output tokens; a near-tie that only bites on very long generated documents.

Claude

128k max output on Fable 5, Mythos 5, Opus 4.8 and Sonnet 5; 64k on Haiku 4.5. A beta header raises it to 300k, but only on the Message Batches API and only for Opus/Sonnet — not the Fable 5 flagship.62The 300k extension is Message Batches API only, and not on Bedrock, Google Cloud or Foundry.

API-level; the 300k cap needs the Batch API and a beta header

ChatGPT

128,000 max output on all three GPT-5.6 models — Sol, Terra and Luna — each with a 1.05M-token context window. No raised batch-only ceiling.64The Batch page only says there's no output-token throughput limit — not a raised per-response ceiling.

API-level; identical across the three GPT-5.6 models

Tie — Flagship to flagship it's 128k vs 128k, and each side has a narrow edge: OpenAI holds 128k down to its cost tier while Claude's Haiku 4.5 stops at 64k, and Anthropic reaches 300k on Opus and Sonnet but only in batch, behind a beta header, and not on its flagship.

Knowledge cutoff / freshness

Flagships are comparable at early-2026; Claude's cheap Haiku tier lags, while ChatGPT keeps its cheap model as fresh as frontier.

Claude

Anthropic publishes two fields: a reliable knowledge cutoff and a broader training-data cutoff. Fable 5, Opus 4.8 and Sonnet 5: Jan 2026 on both. Haiku 4.5: reliable Feb 2025, training Jul 2025.62The two values coincide on the flagships and diverge only on Haiku 4.5 and older models. Month granularity, no day.

Same across API and plans

ChatGPT

16 Feb 2026 for all three GPT-5.6 models — Sol, Terra and Luna (the cost tier) — so the cheap model is as fresh as the frontier. A single, undefined cutoff, no separate training-data range.65Day-level granularity. The same date appears on the index and each model page, but no page defines what it measures.

Same across API tiers

Depends — At the frontier the two are effectively equal — Jan vs 16 Feb 2026, measured with differently-defined yardsticks — so it's a wash for choosing a model; OpenAI's win is at the cost tier, where Luna is current-gen against a small Claude model from Oct 2025.

Frontier API list price per million tokens

For API-metered usage (including Enterprise seat overage), frontier input/output rates are near-identical; Claude's Opus output is cheaper, ChatGPT's Sol slightly pricier out.

Claude

Opus 4.8 (GA frontier): $5 in / $0.50 cache read / $25 out. Fable 5: $10/$1/$50. Sonnet 5: $2/$0.20/$10 introductory through 31 Aug 2026, then $3/$0.30/$15. Batch is 50% off.66Cache reads are 0.1x input on every model. US-only inference adds 1.1x.

Opus 4.8, Fable 5 and Sonnet 5 are GA on the standard API

ChatGPT

GPT-5.6 Sol (frontier): $5 in / $0.50 cached / $30 out. Its siblings: Terra $2.50/$0.25/$15 and Luna $1/$0.10/$6. Batch is 50% off.67Cached input is 0.1x, the same ratio Anthropic uses. Priority is ~2x; data residency adds 10% for models from 5 March 2026.

All listed models on the public pricing page

Depends — On like-for-like standard rates the GA frontier ties on input ($5) and Claude's Opus 4.8 is cheaper on output ($25 vs $30), but OpenAI fields cheaper Sol-generation tiers (Terra, Luna) with no direct counterpart — so which is cheaper turns on the model you run.

Prompt-caching and batch-discount economics

Both cut cost via cache reads (0.1x) and 50% batch discounts; TTLs, cache-write fees and thresholds differ for high-volume workloads.

Claude

Cache read 0.1x input; cache write 1.25x (5-minute TTL) or 2x (1-hour). Minimum cacheable prefix by model: 512 (Fable/Mythos 5), 1,024 (Opus 4.8/Sonnet 5), up to 4,096 (older Opus, Haiku 4.5).68Caching is automatic via a single top-level field, or explicit up to 4 breakpoints. The multipliers stack with Batch.

API pricing; on the Claude API and Claude Platform on AWS

ChatGPT

Caching is on by default for prompts of 1,024 tokens or more, a flat threshold across models. Cached input is a tenth of input. On GPT-5.6+ cache writes cost 1.25x; the only TTL is 30 minutes.69Pre-5.6 models had no write fee and 24-hour retention; on 5.6+ that's gone. Explicit control via prompt_cache_breakpoint.

API pricing; residency endpoints add 10%

Depends — The headline economics are now identical — 0.1x reads and 1.25x writes, both cache automatically; what differs is TTL and the floor: Anthropic sells a 1-hour cache at a 2x write premium where GPT-5.6+ is fixed at 30 minutes, while OpenAI's flat 1,024 floor beats Claude's 4,096 on older Opus and loses to 512 on Fable/Mythos 5.

Fine-tuning / model customization on your data

Neither is a real option — Claude has no first-party fine-tuning and OpenAI is winding its platform down — so plan on prompt/RAG customization.

Claude

No first-party fine-tuning. The only path is third-party: supervised fine-tuning on Amazon Bedrock, and only for Claude 3 Haiku (a March 2024 model) in us-west-2. Anthropic's own story is prompt-side: Agent Skills, caching and memory.70The docs index has no fine-tuning page; the Bedrock table lists exactly one Anthropic row — Claude 3 Haiku.

Nothing on consumer plans; Skills on Pro and up; Bedrock needs an AWS account

ChatGPT

Historically the richer offering — supervised, vision, DPO and reinforcement — but OpenAI is winding it down. Closed to new orgs since 7 May 2026, and all customers stop creating jobs on 6 Jan 2027.71The fine-tunable models are prior-gen (gpt-4.1, gpt-4o, o4-mini); none is current-gen. Tuned models stay available for inference until their base retires.

API platform only; also gated by the org's fine-tuning history

Tie — A company signing up today can't fine-tune either vendor's current model: OpenAI's door shut to new orgs in May 2026 and shuts entirely in January 2027, and Claude's only fine-tunable model is a two-year-old Haiku on Bedrock.

Model deprecation notice period

ChatGPT guarantees ~6 months for GA models vs Claude's 60-day floor — the buffer to re-validate before a model you depend on retires.

Claude

At least 60 days' notice before retirement; each active model also carries a published not-sooner-than floor date, about 12 months out at launch.72The dates apply only to Anthropic-operated platforms (Claude API, AWS, Foundry); Bedrock and Vertex set their own schedules.

Same policy for all API customers

ChatGPT

At least 6 months' notice for GA models, 3 months for specialized variants, and as little as 2 weeks for anything named preview; no published floor per snapshot.73In practice the dates held: a 20 Jul 2026 removal of legacy audio models was set 6 months out. No minimum lifetime is published for a pinned snapshot.

Policy is API-wide; no tier-specific extension

ChatGPT — Six months beats sixty days threefold, and both vendors have honoured their stated windows — though Anthropic is the only one that tells you a model's earliest possible retirement date on the day you adopt it.

If you must pick one

There's no overall winner. But if you need a rule of thumb, here's how we choose.

  • If your data must stay in the EU — ChatGPT is the only one that runs inference in the EU; Claude runs in the US only.
  • If you need to generate images or video, or show a screen in a voice chat — that's ChatGPT; Claude simply doesn't do those.
  • If you buy through an existing cloud (AWS, Google Cloud, Azure), need audit logs kept for years, or indemnity that also covers the entry-level business plan — Claude.
  • If the work is mostly chart-heavy documents or connecting to internal systems with no gatekeeper — Claude; if you need a public app store or research you can export to Word — ChatGPT.

Frequently asked

Which is better for the office — ChatGPT or Claude?

There's no single answer. For most offices both work, and the call depends on the task: images and video lean ChatGPT, chart-heavy documents and connecting to internal systems lean Claude. The table above breaks it down by layer.

Is our data used to train the models?

On both vendors' business tiers — not by default. With Claude it's written into the Commercial Terms; with ChatGPT it's the stated position. The specifics, with a source link, are in the admin and security layer.

Can we connect the tools to our internal systems?

Yes, on both — both are built on MCP. Claude accepts any remote MCP server by pasting a URL; ChatGPT publishes apps through a review portal. Fast reach vs a curated catalogue.

How much does it cost?

A paid individual plan is $20/month on both; Claude adds a discounted annual option. A business seat is identical to the dollar — $20 annual, $25 monthly. Full detail in the plans and pricing layer.

Which is better for Israel?

Both are sold in Israel and neither localises its interface into Hebrew. The one practical difference: ChatGPT lists the shekel (₪) as a supported billing currency, and Claude publishes no currency list.

Sources

Every figure on this page was checked against an official vendor page — anthropic.com, openai.com and their docs. The page is kept current; the date above is the last check.

  1. support.claude.com support.claude.com/en/articles/8241126-upload-files-to-claude April 22, 2026 (Last Updated, shown on page)
  2. help.openai.com help.openai.com/en/articles/10416312-visual-retrieval-with-pdfs-faq Updated ~14 Jul 2026 (page shows "7 days ago" as of 21 Jul 2026)
  3. support.claude.com support.claude.com/en/articles/11088861-use-research-on-claude June 2, 2026
  4. help.openai.com help.openai.com/en/articles/10500283-deep-research-in-chatgpt Updated 7 days ago as shown on 21 Jul 2026, i.e. ~14 Jul 2026
  5. support.claude.com support.claude.com/en/articles/9002504-can-claude-produce-images March 16, 2026
  6. help.openai.com help.openai.com/en/articles/11084440-chatgpt-images-faq Updated ~18 Jul 2026
  7. support.claude.com support.claude.com/en/articles/12012173-get-started-with-claude-in-chrome undated; page shows "Updated this week" as of 21 Jul 2026
  8. help.openai.com help.openai.com/en/articles/6825453-chatgpt-release-notes Page "Updated: 3 days ago" (newest entry 16 Jul 2026); entries cited: 9 Jul 2026 (Atlas retirement, desktop built-in browser), 18 Jun 2026 (Record & Replay), 17 Jun 2026 (scheduled tasks), 11 Jun 2026 (Codex Developer mode), 4 Jun 2026 (Lockdown Mode)
  9. support.claude.com support.claude.com/en/articles/12111783-create-and-edit-files-with-claude April 29, 2026
  10. help.openai.com help.openai.com/en/articles/8555545-file-uploads-faq updated 8 days ago (~13 July 2026)
  11. developers.openai.com developers.openai.com/api/docs/guides/video-generation undated (deprecation notice dated 2026-03-24 per https://developers.openai.com/api/docs/deprecations)
  12. support.claude.com support.claude.com/en/articles/11101966-use-voice-mode 21 May 2026 (last updated)
  13. help.openai.com help.openai.com/en/articles/8400625-voice-mode-faq undated; video/screen-share rollout originally announced Dec 2024, corroborated via search index fetched 22 Jul 2026
  14. support.claude.com support.claude.com/en/articles/11175166-about-custom-connectors-remote-mcp April 2, 2026
  15. developers.openai.com developers.openai.com/apps-sdk/ undated
  16. claude.com claude.com/pricing undated (fetched 22 Jul 2026)
  17. chatgpt.com chatgpt.com/pricing undated (fetched 22 Jul 2026); help.openai.com Free Tier FAQ updated ~15 Jul 2026 — https://help.openai.com/en/articles/9275245-chatgpt-free-tier-faq
  18. support.claude.com support.claude.com/en/articles/9517075-what-are-projects 16 March 2026
  19. help.openai.com help.openai.com/en/articles/10169521-projects-in-chatgpt 15 July 2026
  20. support.claude.com support.claude.com/en/articles/9519177-how-can-i-create-and-manage-projects updated ~2026-07 (page shows 'updated this week')
  21. help.openai.com help.openai.com/en/articles/8554407-gpts-faq undated (not captured; direct fetch blocked)
  22. support.claude.com support.claude.com/en/articles/9487310-what-are-artifacts-and-how-do-i-use-them undated (article shows 'Updated today'); mobile rollout dated July 21, 2025 per claude.com/blog/build-artifacts
  23. help.openai.com help.openai.com/en/articles/9930697-what-is-the-canvas-feature-in-chatgpt-and-how-do-i-use-it academy.openai.com page: updated 2026-05-29 (orig. 2025-09-08); help article undated, mobile-gap wording reconfirmed 22 Jul 2026
  24. support.claude.com support.claude.com/en/articles/13854387-schedule-recurring-tasks-in-claude-cowork undated (help page shows 'Updated over a week ago')
  25. help.openai.com help.openai.com/en/articles/10291617-scheduled-tasks-in-chatgpt undated
  26. support.claude.com support.claude.com/en/articles/10593882-share-and-unshare-chats 15 June 2026
  27. help.openai.com help.openai.com/en/articles/8798634-chatgpt-shared-links-faq-business-version undated
  28. support.claude.com support.claude.com/en/articles/13133195-set-up-jit-or-scim-provisioning undated (page shows 'Updated over 3 weeks ago' as of 2026-07-22)
  29. help.openai.com help.openai.com/en/articles/10011769-scim-integration-faq 2026-07-15 (page shows 'Updated: 7 days ago' as of 2026-07-22)
  30. platform.claude.com platform.claude.com/docs/en/manage-claude/compliance-integration-patterns undated
  31. help.openai.com help.openai.com/en/articles/9261474-openai-compliance-platform-for-enterprise-and-edu-customers 14 July 2026
  32. support.claude.com support.claude.com/en/articles/9267276-roles-and-permissions 18 June 2026
  33. help.openai.com help.openai.com/en/articles/8542216-managing-members-seat-types-and-roles-in-chatgpt-business undated (not directly rendered; role definitions corroborated via search index, fetched 22 Jul 2026)
  34. support.claude.com support.claude.com/en/articles/10440198-configure-custom-data-retention-controls-for-enterprise-plans 16 March 2026
  35. openai.com openai.com/enterprise-privacy/ 8 January 2026
  36. support.claude.com support.claude.com/en/articles/10015870-what-certifications-has-anthropic-obtained 16 March 2026
  37. trust.openai.com trust.openai.com undated (SOC 2 period Jan 1–Jun 30 2025)
  38. support.claude.com support.claude.com/en/articles/13132885-set-up-single-sign-on-sso undated — page displays only the relative marker "Updated over 3 weeks ago" (checked 2026-07-21)
  39. help.openai.com help.openai.com/en/articles/10468051-sso-overview undated — page displays only "Updated: 7 days ago", i.e. approximately 2026-07-14 (read in a browser session on 2026-07-21; the URL returns HTTP 403 to automated fetchers)
  40. privacy.claude.com privacy.claude.com/en/articles/7996868-is-my-data-used-for-model-training Last updated March 16, 2026 (Commercial Terms quotes: https://www.anthropic.com/legal/commercial-terms, effective June 17, 2025)
  41. developers.openai.com developers.openai.com/api/docs/guides/your-data undated (redirect target of platform.openai.com/docs/guides/your-data; fetched 2026-07-21)
  42. support.claude.com support.claude.com/en/articles/13200993-restrict-access-to-claude-with-ip-allowlisting March 16, 2026
  43. help.openai.com help.openai.com/en/articles/12111596-ip-allowlisting-for-chatgpt Updated ~2026-07-14
  44. platform.claude.com platform.claude.com/docs/en/build-with-claude/claude-in-microsoft-foundry undated
  45. learn.microsoft.com learn.microsoft.com/en-us/azure/ai-foundry/foundry-models/concepts/models-sold-directly-by-azure 7 July 2026
  46. support.claude.com support.claude.com/en/articles/11175166-about-custom-connectors-remote-mcp-servers 2 April 2026
  47. developers.openai.com developers.openai.com/apps-sdk/deploy/connect-chatgpt undated (sibling API page dated 17 December 2025 for the connectors-to-apps rename)
  48. platform.claude.com platform.claude.com/docs/en/api/service-tiers undated (fetched 2026-07-22); status figures fetched 2026-07-22; Commercial Terms effective 2025-06-17
  49. developers.openai.com developers.openai.com/api/docs/guides/priority-processing undated (fetched 2026-07-22); status figures cover Apr 2026–Jul 2026, fetched 2026-07-22
  50. code.claude.com code.claude.com/docs/en/sandboxing undated
  51. learn.chatgpt.com learn.chatgpt.com/docs/sandboxing undated
  52. platform.claude.com platform.claude.com/docs/en/build-with-claude/data-residency undated
  53. www.anthropic.com www.anthropic.com/supported-countries undated (© 2026)
  54. help.openai.com help.openai.com/en/articles/10421635-multicurrency-billing updated ~15 Jul 2026 ("7 days ago")
  55. www.anthropic.com www.anthropic.com/legal/commercial-terms Effective June 17, 2025
  56. openai.com openai.com/policies/service-terms/ Updated June 12, 2026 (Service Terms); Business Terms updated December 1, 2025, effective January 1, 2026
  57. help.openai.com help.openai.com/en/articles/8792828-what-is-chatgpt-business 2026-07-20 (page shows "Updated: 23 hours ago")
  58. help.openai.com help.openai.com/en/articles/9793128-about-chatgpt-pro-tiers page shows a relative "Updated: 7 days ago"; no absolute date is printed, so this resolves to approx. 2026-07-14 as read on 2026-07-21
  59. support.claude.com support.claude.com/en/articles/11049741-what-is-the-max-plan undated (header shows only "Updated over a week ago"); fetched 2026-07-21. Corroborated by https://claude.com/pricing (undated, fetched 2026-07-21).
  60. learn.chatgpt.com learn.chatgpt.com/docs/pricing undated (fetched 2026-07-21; page shows only an unrelated June 11–24, 2026 promo window)
  61. code.claude.com code.claude.com/docs/en/costs undated (corroborated against https://claude.com/pricing, undated, and https://support.claude.com/en/articles/9797557-usage-limit-best-practices, updated 2 Jun 2026; all three re-fetched 2026-07-21)
  62. platform.claude.com platform.claude.com/docs/en/about-claude/models/overview undated
  63. developers.openai.com developers.openai.com/api/docs/models/all undated
  64. developers.openai.com developers.openai.com/api/docs/models undated (fetched 2026-07-21)
  65. developers.openai.com developers.openai.com/api/docs/models/gpt-5.6-luna undated
  66. platform.claude.com platform.claude.com/docs/en/about-claude/pricing undated
  67. developers.openai.com developers.openai.com/api/docs/pricing undated
  68. platform.claude.com platform.claude.com/docs/en/about-claude/pricing#prompt-caching undated
  69. developers.openai.com developers.openai.com/api/docs/guides/prompt-caching undated
  70. docs.aws.amazon.com docs.aws.amazon.com/bedrock/latest/userguide/custom-model-fine-tuning.html undated
  71. developers.openai.com developers.openai.com/api/docs/guides/model-optimization undated (deprecation dates on the page run to 6 January 2027)
  72. platform.claude.com platform.claude.com/docs/en/about-claude/model-deprecations undated (page current as of 22 Jul 2026; most recent entry 2026-06-05)
  73. developers.openai.com developers.openai.com/api/docs/deprecations undated (most recent entry 2026-07-20)
Back to the home page