Locul / Tools / AI memory and context limits

Reference table · checked 7 Oct 2026

What ChatGPT, Claude, Gemini, Copilot and Perplexity remember

One table of what the five big assistants keep about you: saved memories, past-chat recall, projects, instructions, context windows, file limits, retention, training and export. Every cell links to the vendor's own documentation and says when it was checked.

150cells checked
88vendor pages cited
15marked not documented
Monthlyre-check

The short answer

AI memory limits depend more on your plan than on the brand. ChatGPT, Claude, Copilot and Perplexity keep saved memories on their free plans, Gemini saves explicit facts only on a paid plan in the US, and context windows run from 27K tokens (ChatGPT Free) to 1M (Claude, Gemini Pro). On business plans, an admin usually decides.

We checked all 150 cells against 88 vendor pages on 7 October 2026. Fifteen are not documented by the vendor at all. Six findings stand out:

  1. Context windows range from 27K to 1M tokens on free plans. ChatGPT Free gives 27K with its Instant model, Gemini's free app 32K, and Claude "up to 1M, varies by model". See the row.
  2. Two of five publish no context window at all. Microsoft and Perplexity do not state a number for any plan, so any figure you see quoted for them is a guess.
  3. Personal plans train on your chats unless you opt out at OpenAI, Anthropic, Google and Perplexity. Microsoft says the new Copilot app does not use prompts to train foundation models. All five business tiers default to no training.
  4. Deleting a chat does not always delete the memory. Claude and Microsoft's docs say memories made from a chat survive its deletion; Gemini says the reverse: to make it forget, delete the chats.
  5. "Deleted" usually means up to 30 days. ChatGPT and Claude remove deleted chats within 30 days, and Perplexity deletes an account's data 30 days after you close it. If you let Anthropic train on your chats, it may keep de-identified copies for up to 5 years.
  6. Every assistant has a memory-off mode, but none is instant: Perplexity's Incognito threads expire in 24 hours, Gemini keeps temporary chats 72 hours, ChatGPT and Claude up to 30 days.

In the vendors' own words

"When a conversation expires or is deleted, related memory entries generated from it won't be removed, but you can delete individual memories at any time."
Anthropic, Claude Help Center
"To delete something that Gemini remembered about you, delete all chats with this info from Gemini Apps Activity."
Google, Gemini Apps Help
"The portion available for user input is smaller than the total window, as space is also used for system instructions (including tools and personality), memories (if enabled), and internal processing..."
OpenAI, ChatGPT pricing page
"Prompts, responses, and your file contents when using the Microsoft Copilot app aren't used to train foundation models."
Microsoft Support, Copilot privacy
The table

What does each AI assistant remember, by plan?

Each assistant can remember you in five ways (saved memories, past chats, projects, standing instructions and the open chat itself), and your plan changes most of them. Each cell gives the short answer, the detail, and the vendor page it comes from. Where a vendor's own documentation does not say, the cell says Not documented instead of guessing. Filter by plan, hide assistants you do not use, or compare two side by side.

Plan
Show

Showing the short answer for every plan. Pick Free, Paid or Team to read the detail in each cell.

Yes documented and available Opt-in limited, off by default or admin-controlled No not offered 128K a documented number Not documented the vendor does not say
Which plans count as Free, Paid and Team for each assistant
ChatGPTFree FreePaid Go, Plus, ProTeam Business, Enterprise, Edu
ClaudeFree FreePaid Pro, MaxTeam Team, Enterprise
GeminiFree Free (no Google AI plan)Paid Google AI Plus, Pro, UltraTeam Workspace business, enterprise and education
CopilotFree Copilot app (free)Paid Microsoft 365 Personal, Family, Premium, ProTeam Microsoft Copilot and Copilot Chat (work)
PerplexityFree FreePaid Pro, MaxTeam Enterprise Pro, Enterprise Max
What ChatGPT, Claude, Gemini, Copilot and Perplexity remember, by plan, each cell linked to the vendor's documentation, checked 7 Oct 2026
What it remembersChatGPTClaudeGeminiCopilotPerplexity
Saved memoriesDoes it keep facts about you from one chat to the next?What this means: Things you said weeks ago can shape today's answer. Open the memory list once in a while: what is stored there is what it "knows" about you.ChatGPT
  • FreeLimitedThe plan page lists Memory as "Limited" on Free. Saved memories can be switched off in Settings. No memory capacity is published.
  • PaidYesGo: Memory "Yes". Plus and Pro: "Expanded". You can review, change or remove what it remembers. No capacity number published.
  • TeamYesBusiness, Enterprise and Edu have memory. Enterprise and Edu users can correct memory, turn it off or go back to legacy saved memories.
ChatGPT pricing: compare plans, Memory and new controls for ChatGPT, Memory in ChatGPT, Memory FAQ (Business Version), ChatGPT Enterprise and Edu release notes · checked 7 Oct 2026
Claude
  • FreeOn by defaultMemory from chats is on by default; view, edit or delete each memory topic in Settings > Memory. No capacity limit documented.
  • PaidOn by defaultMemory from chats is on by default; view, edit or delete each memory topic in Settings > Memory. No capacity limit documented. On by default, topics editable and deletable, Pause or Reset memory available.
  • TeamAdmin controlsOwners must enable memory for the org; then off for each member until they turn it on. Not for HIPAA or custom-retention orgs.
Use Claude's chat search and memory to build on previous context, Claude pricing: Compare features across plans · checked 7 Oct 2026
Gemini
  • FreeNoAsking Gemini to remember specific facts needs a Google AI plan, US only. Free users still get Memory from past chats.
  • PaidYes (US only)Say 'Remember that...' to save facts; correct them in chat. Requires 18+, personal account, US, Memory on. No capacity limit documented.
  • TeamNoNot available on work or school accounts.
Manage what Gemini remembers about you and customize responses, Get personalization with memory of your past Gemini chats · checked 7 Oct 2026
Copilot
  • FreeOn by defaultCopilot Memory saves key details; Saved memories toggle; delete one or all in Settings > Personalization. No capacity limit stated.
  • PaidOn by defaultSame Copilot app memory for Microsoft 365 Personal, Family, Premium, Pro subscribers. No capacity limit stated.
  • TeamOn by defaultCopilot memory for Copilot Chat users with or without a license (preview). Admin Enhanced personalization control is on by default.
Microsoft Copilot for individuals: your privacy controls and..., Manage Copilot Memory in Microsoft Copilot, Manage Copilot personalization and memory (Microsoft Learn), Microsoft Privacy Statement (Last Updated: September 2026), A... · checked 7 Oct 2026
Perplexity
  • FreeYesMemory is described with no plan restriction; view, delete or clear all in Settings > Personalize. Default state and capacity not documented.
  • PaidYesSame Memory feature for Pro and Max. Max users of Computer also get Brain, a memory system in research preview.
  • TeamAdmin controlsAdmins enable "Reference saved memories" org-wide; users can opt out. Memories belong to the org, not the user.
Memory, Which Perplexity Subscription Plan is right for you?, Memory for Enterprise Organizations · checked 7 Oct 2026
Past-chat recallCan a new chat draw on your old conversations?What this means: Separate from saved memories. Even with nothing saved, the assistant may pull details from earlier chats, and it has its own switch.ChatGPT
  • FreeLimited"Memory with past chats: Limited". OpenAI calls it a lightweight version that gives short-term continuity across chats. Can be turned off.
  • PaidYesGo: "Yes". Plus and Pro: "Expanded", a longer-term understanding of you. Turn off "chat history" referencing in Settings.
  • TeamYesWith memory on, ChatGPT on Business, Enterprise and Edu can use relevant context from past chats.
ChatGPT pricing: compare plans, Memory and new controls for ChatGPT, ChatGPT Business release notes, ChatGPT Enterprise and Edu release notes · checked 7 Oct 2026
Claude
  • FreeNoSearch and reference past chats is documented for paid plans only.
  • PaidOn by defaultPro and Max: Claude can search past chats (RAG tool call); on by default once rolled out, toggle "Search and reference chats".
  • TeamYesTeam and Enterprise included; not available if an Enterprise org uses customer-managed encryption keys.
Use Claude's chat search and memory to build on previous context · checked 7 Oct 2026
Gemini
  • FreeYes'Memory' of past chats personalizes replies. Needs 18+, personal account, Keep Activity on. Default state not documented.
  • PaidYesSame Memory feature as free (Settings > Personal Intelligence > Memory). Not available in Gems or Live chats.
  • TeamNoMemory of past chats and Personal Intelligence are not available for work or school accounts.
Get personalization with memory of your past Gemini chats, Generative AI in Google Workspace Privacy Hub · checked 7 Oct 2026
Copilot
  • FreeYesCopilot infers memories from past chats via the Chat history toggle; turning it off deletes inferences after 30 days.
  • PaidYesCopilot infers memories from past chats via the Chat history toggle; turning it off deletes inferences after 30 days.
  • TeamOn by defaultChat history (plus work insights with a work account) personalization; admin Enhanced personalization control is on by default.
How Microsoft Copilot Chat history works, Microsoft Copilot for individuals: your privacy controls and..., Manage Copilot personalization and memory (Microsoft Learn) · checked 7 Oct 2026
Perplexity
  • FreeYesMemory can reference past searches ("Search history"), with its own toggle in Personalize. Plan limits and default not documented.
  • PaidYesSame for Pro and Max: past questions and answers can be referenced and cited; separate toggle from saved memories.
  • TeamAdmin controlsAdmin setting "Reference search library"; if disabled org-wide it is off for all users.
Memory, Memory for Enterprise Organizations · checked 7 Oct 2026
Projects and workspacesIs there a space with its own files, instructions and memory?What this means: A project keeps the files and rules for one piece of work, so you stop re-uploading the same documents. Limits apply per project.ChatGPT
  • FreeYes, 5 filesUnlimited projects; up to 5 files per project.
  • Paid25 or 40 filesGo and Plus: 25 files per project. Pro: 40 files per project. Unlimited projects.
  • Team40 filesBusiness, Enterprise, Edu: up to 40 files per project. Unlimited projects.
ChatGPT pricing: compare plans, Projects in ChatGPT, Uploading files and audio to ChatGPT · checked 7 Oct 2026
Claude
  • FreeUp to 5Free users can create a maximum of five projects, each with its own knowledge, instructions and separate project memory.
  • PaidUnlimitedUnlimited projects; RAG expands project knowledge up to 10x once it nears the context window. Project files 30MB each.
  • TeamYesSame plus sharing with org members (Can view / Can edit) and admin sharing controls.
What are projects?, Use Claude's chat search and memory to build on previous context, Retrieval augmented generation (RAG) for projects, Upload files to Claude, Claude pricing: Compare features across plans · checked 7 Oct 2026
Gemini
  • FreeGems + NotebooksGems on all plans (becoming skills Nov 2026). Gemini Notebook in Gemini: 100 notebooks, 50 sources each.
  • PaidGems + NotebooksNotebooks: Plus 200 x 100 sources; Pro 500 x 300; Ultra 500 x 500 (20 TB) or 600 (30 TB). Gem file caps not documented.
  • TeamGems + NotebooksGems remain until March 2027 (business) or June 2027 (education). Gemini Notebook is a core Workspace service. Limits not documented.
Gemini Apps limits & upgrades for Google AI subscribers, Use Gems in Gemini Apps, About the transition from Gems to skills, Notebooks in Gemini Apps, Learn about Gemini Notebook's plans, Generative AI in Google Workspace Privacy Hub · checked 7 Oct 2026
Copilot
  • FreeNot documentedCopilot Notebooks availability for the free consumer app is not stated.
  • PaidNotebooksCopilot Notebooks for Microsoft 365 Personal, Family, Premium, Pro. Reference cap for home plans not stated.
  • TeamNotebooks + PagesCopilot Notebooks: licensed users ground on up to 300 references; Copilot Chat users up to 50.
Frequently asked questions about Microsoft Copilot Notebooks, Add references to your Microsoft Copilot Notebook, Provide custom instructions for your Microsoft Copilot Notebook · checked 7 Oct 2026
Perplexity
  • FreeNot documentedSpaces are documented, but no Free plan availability or file limit is stated.
  • PaidYesSpaces with custom instructions and files: Pro up to 50 files per Space, Max up to 500; 50 MB per file; up to 5 contributors.
  • TeamYesEnterprise Pro: 500 files per Space; Enterprise Max: 5,000. Org file repository 500 files. Spaces do not use memories.
What are Spaces?, Enterprise file limits, Memory for Enterprise Organizations · checked 7 Oct 2026
Custom instructionsCan you set standing instructions for every chat?What this means: The one memory you fully control: text you write once and the assistant reads every time. Keep it short, because it uses up context.ChatGPT
  • FreeYesSettings > Customize ChatGPT > Custom Instructions. No character limit on the pages we could read.
  • PaidYesSettings > Customize ChatGPT > Custom Instructions. No character limit on the pages we could read.
  • TeamYesBusiness can set custom instructions and a personality; they carry over when you move a personal workspace into Business.
ChatGPT Custom Instructions, Memory in ChatGPT, ChatGPT Business release notes, Migrating your personal workspace to ChatGPT Business or Ente... · checked 7 Oct 2026
Claude
  • FreeYes"Instructions for Claude" in Settings apply to all chats. Character limit not documented.
  • PaidYesSame account-wide Instructions for Claude, plus per-project instructions. Character limit not documented.
  • TeamYesUser instructions plus owner-set organization instructions (max 3,000 characters), which take precedence.
Understanding Claude's personalization features, Claude pricing: Compare features across plans, Set organization instructions · checked 7 Oct 2026
Gemini
  • FreeYes'Instructions for Gemini' ('Saved info' in some regions) apply to every chat. Character limit not documented.
  • PaidYesSame Instructions for Gemini, plus skills (18+): up to 100 active at a time. Character limit not documented.
  • TeamNoPersonalization features, including instructions, are not available on work or school accounts.
Manage what Gemini remembers about you and customize responses, Gemini Apps Privacy Hub, About the transition from Gems to skills, Get personalization in Gemini Apps · checked 7 Oct 2026
Copilot
  • FreeYesCustom instructions in Settings > Personalization (shared Copilot help article). Character limit not stated.
  • PaidYesCustom instructions in Settings > Personalization (shared Copilot help article). Character limit not stated. Character limit not stated.
  • TeamYesCustom instructions; admins can disable via Enhanced personalization. Not discoverable by admins; user can export. No character limit stated.
Customize how Microsoft Copilot responds to you, Personalize what Microsoft Copilot remembers, Manage Copilot personalization and memory (Microsoft Learn) · checked 7 Oct 2026
Perplexity
  • FreeYesSettings > Personalization has a free-text "Custom instructions" field plus occupation, response length. No character limit documented.
  • PaidYesSame account personalization, plus per-Space custom instructions. No character limit documented.
  • TeamYesSame personalization and Space instructions; no org-wide instruction feature or character limit documented.
Account Settings, What are Spaces? · checked 7 Oct 2026
Context windowHow much text one chat can hold at onceWhat this means: Past this size, the start of a long chat or a big file drops out of view. A bigger window means longer documents before it forgets.ChatGPT
  • Free27K27K tokens in total with the Instant model; "Varies" with the reasoning model. Instructions and memories use part of it.
  • Paid54K to 400KGo and Plus: 54K (Instant), 256K (Reasoning). Pro: 128K (Instant), 400K (Reasoning). Totals, not just your input.
  • TeamNot documentedThe business plan pages we could read give no number. An undated Enterprise file guide, seen only in search results, mentions 128K.
ChatGPT pricing: compare plans, Optimizing File Uploads in ChatGPT Enterprise · checked 7 Oct 2026
Claude
  • FreeUp to 1MPricing page: "Up to 1M varies by model". No per-model table for Free in the help center.
  • Paid200K to 1MChat: 1M on Fable 5.1, Opus 5.5, Opus 5, Sonnet 5.5, Sonnet 5; 500K on Fable 5, Opus 4.6 to 4.8, Sonnet 4.6; 200K others.
  • Team200K to 1MSame per-model table covers Team and Enterprise; pricing page shows "Up to 1M varies by model".
Claude pricing: Compare features across plans, How large is the context window on paid Claude plans?, How do usage and length limits work? · checked 7 Oct 2026
Gemini
  • Free32KWithout an AI plan: 32k tokens.
  • Paid128K / 1MAI Plus: 128k tokens. AI Pro and AI Ultra: 1 million tokens. Deep Think: 192 thousand tokens.
  • Team32K / 1M32 thousand for Gemini as an additional service and Business Starter/Essentials tier; 1 million for Business Standard/Plus, Enterprise and AI add-ons.
Gemini Apps limits & upgrades for Google AI subscribers, Upload & analyze files in Gemini Apps, Use Gemini Apps with a work or school Google Account · checked 7 Oct 2026
Copilot
  • FreeNot documentedNo token context window is documented for the Copilot app.
  • PaidNot documentedNo token context window is documented for Microsoft 365 home plans.
  • TeamNot documentedNo token context window is documented for Microsoft Copilot or Copilot Chat.
No vendor page states this · checked 7 Oct 2026
Perplexity
  • FreeNot documentedNo token context window is stated for the Perplexity app.
  • PaidNot documentedNo token context window is stated for the app; long files are summarized to their most important parts.
  • TeamNot documentedNo token context window is stated for Enterprise Pro or Enterprise Max.
No vendor page states this · checked 7 Oct 2026
File uploadsHow many files, and how big, per chatWhat this means: Decides whether a whole report or codebase fits in one go, or whether you have to split it into parts.ChatGPT
  • Free512 MB per file512 MB hard limit per file. Uploads are "Limited" on Free; up to 5 files per project. No daily cap on the pages we could read.
  • Paid512 MB per file512 MB per file. Projects: Go and Plus 25 files, Pro 40 files. No daily cap on the pages we could read.
  • Team512 MB per file512 MB per file. Business, Enterprise and Edu: up to 40 files per project.
ChatGPT pricing: compare plans, File storage and Library in ChatGPT, Uploading files and audio to ChatGPT · checked 7 Oct 2026
Claude
  • Free500MB, 20 filesChat: 500MB per file, up to 20 files per chat, PDFs up to 1000 pages. Project files: 30MB each. No per-plan split documented.
  • Paid500MB, 20 filesSame documented limits: 500MB per file, 20 files per chat, 8000x8000 px images; project files 30MB each.
  • Team500MB, 20 filesSame documented limits; no separate business-tier file caps documented. Daily upload caps not documented.
Upload files to Claude · checked 7 Oct 2026
Gemini
  • Free10 files, 100 MBUp to 10 files per prompt; 100 MB per file, 2 GB per video; 5 min total video, 10 min audio; rolling limits on chats with files.
  • Paid10 files, 100 MBSame per-file caps; Pro and Ultra extend video to 1 hour and audio to 3 hours and get higher upload limits.
  • Team10 files, 100 MBSame upload article applies; Drive uploads need admin to enable apps. Per-edition upload caps not documented.
Upload & analyze files in Gemini Apps · checked 7 Oct 2026
Copilot
  • FreeNot documentedFile upload size or count limits are not stated for the consumer app.
  • PaidNot documentedFile upload size or count limits are not stated for home plans.
  • TeamNot documentedMicrosoft says upload limits apply but gives no size or count. Uploaded files are stored in OneDrive for Business.
Frequently asked questions about Microsoft Copilot Chat (Micr..., Add references to your Microsoft Copilot Notebook · checked 7 Oct 2026
Perplexity
  • FreeLimited"Basic file uploads (limited)"; 50 MB per file. Daily or per-thread counts for Free not documented.
  • Paid50 MB50 MB per file; Pro has increased upload rate limits; Spaces: Pro 50 files, Max 500. Exact weekly counts not documented.
  • Team30 files per thread30 files (50MB each) per thread upload, 4 images per query; 100/week Enterprise Pro, 1,000/week Enterprise Max.
Which Perplexity Subscription Plan is right for you?, File Uploads, What are Spaces?, Enterprise file limits · checked 7 Oct 2026
RetentionHow long chats are kept, and how long after you delete themWhat this means: Deleting a chat hides it at once, but a copy can stay on the vendor's servers for a set time. This row says how long.ChatGPT
  • Free30 days after deleteDeleted chats and data are removed within 30 days unless kept for safety or legal reasons. Temporary Chats auto-delete within 30 days.
  • Paid30 days after deleteDeleted chats and data are removed within 30 days unless kept for safety or legal reasons. Temporary Chats auto-delete within 30 days.
  • TeamAdmin controlsWorkspace admins set how long data is kept. Deleted or unsaved conversations are removed within 30 days unless the law requires longer.
OpenAI Privacy Policy, Enterprise privacy at OpenAI · checked 7 Oct 2026
Claude
  • Free30 days or 5 yearsDeleted chats purged from backend within 30 days. If training is allowed, de-identified data kept up to 5 years. Incognito: 30 days.
  • Paid30 days or 5 yearsDeleted chats purged from backend within 30 days. If training is allowed, de-identified data kept up to 5 years. Incognito: 30 days. Memory follows chat retention; deleting a chat does not remove memories made from it.
  • TeamAdmin controlsDeleted chats purged within 30 days; Enterprise owners can set custom retention (minimum 30 days). Default: kept indefinitely.
How long do you store my data?, Updates to Consumer Terms and Privacy Policy, Use incognito chats, Use Claude's chat search and memory to build on previous context, How long do you store my organization’s data?, Configure custom data retention controls for Enterprise plans · checked 7 Oct 2026
Gemini
  • Free18 monthsActivity auto-deletes after 18 months by default (3, 36 months or never). Keep Activity off or temporary chats: 72 hours. Reviewed chats: up to 3 years.
  • Paid18 monthsActivity auto-deletes after 18 months by default (3, 36 months or never). Keep Activity off or temporary chats: 72 hours. Reviewed chats: up to 3 years. Instructions stay until you delete them.
  • TeamAdmin controlsDefault 18 months; admins choose 3, 18 or 36 months or turn history off (then 72 hours). Vault rules can apply.
Manage & delete your activity in Gemini Apps, Gemini Apps Privacy Hub, Generative AI in Google Workspace Privacy Hub, Control temporary chats and chat deletion in the Gemini App · checked 7 Oct 2026
Copilot
  • FreeNot documentedThe new Copilot app (from 18 Aug 2026) states no period. The older app's FAQ said 18 months. You can delete chats at any time.
  • PaidNot documentedThe new Copilot app (from 18 Aug 2026) states no period for Microsoft 365 home plans either. You can delete chats at any time.
  • TeamAdmin controlsAdmins set Purview retention for Copilot interactions; Purview policies do not apply to memory. Saved memories kept until user deletes.
Privacy FAQ for Microsoft Copilot (older app version), Data, Privacy, and Security for Microsoft Copilot (Microsoft..., Manage Copilot personalization and memory (Microsoft Learn) · checked 7 Oct 2026
Perplexity
  • FreeUntil deletedKept while account is active; account deletion completes in 30 days. Thread files 30 days (one article says 90). Incognito 24 hours.
  • PaidUntil deletedKept while account is active; account deletion completes in 30 days. Thread files 30 days (one article says 90). Incognito 24 hours. Deleted memories may be logged up to 30 days. Logged-out threads expire after 14 days.
  • TeamAdmin controlsThread files deleted after 7 days. Custom thread retention for orgs with 50+ seats or 1 Enterprise Max seat.
How long does Perplexity retain my search history, profile da..., What’s the difference between typing a question and uploading..., Security and Privacy with File Uploads, Memory, Where did my threads go?, Data Retention for Enterprise, Memory for Enterprise Organizations · checked 7 Oct 2026
Training on your chatsAre your chats used to train models by default?What this means: The default is what happens if you never open settings. Business plans usually start opted out; personal plans often start opted in.ChatGPT
  • FreeYes, opt-outChats on personal plans may be used to train models. Turn off "Improve the model for everyone" in Data Controls; it applies to new chats.
  • PaidYes, opt-outChats on personal plans may be used to train models. Turn off "Improve the model for everyone" in Data Controls; it applies to new chats. The plan page says "Opt-out available".
  • TeamNoBusiness, Enterprise and Edu data is not used for training by default.
ChatGPT pricing: compare plans, How your data is used to improve model performance, Enterprise privacy at OpenAI, File storage and Library in ChatGPT · checked 7 Oct 2026
Claude
  • FreeYes, opt-outUser choice (asked at signup); training on unless you turn off "Help improve our AI models" in Settings > Privacy. Incognito never used.
  • PaidYes, opt-outUser choice (asked at signup); training on unless you turn off "Help improve our AI models" in Settings > Privacy. Incognito never used. Toggle "Help improve our AI models" under Settings > Privacy.
  • TeamNoNo training on Team or Enterprise content by default, unless you send thumbs up/down feedback or otherwise allow it.
Claude pricing: Compare features across plans, Privacy Policy (effective September 10, 2026), Updates to Consumer Terms and Privacy Policy, How do I change my model improvement privacy settings?, How long do you store my data?, Is my data used for model training? (Commercial) · checked 7 Oct 2026
Gemini
  • FreeYes, opt-outKeep Activity is on by default (18+) and lets Google use chats to improve services, including training, with human review. Turn off Keep Activity to opt out.
  • PaidYes, opt-outKeep Activity is on by default (18+) and lets Google use chats to improve services, including training, with human review. Turn off Keep Activity to opt out. Governed by the Keep Activity setting in Gemini Apps Activity. Temporary chats are not used for training.
  • TeamNoCore-service Workspace editions: chats and files not human reviewed or used to train models. Gemini as an additional service: may be used.
Manage & delete your activity in Gemini Apps, Gemini Apps Privacy Hub, Use Gemini Apps with a work or school Google Account · checked 7 Oct 2026
Copilot
  • FreeNoNew Copilot app: prompts, responses and file contents are not used to train foundation models. Older app used them unless you opted out.
  • PaidNoSame for the Copilot app and for Copilot in Microsoft 365 apps for home.
  • TeamNoPrompts, responses and Microsoft Graph data are not used to train foundation LLMs.
Microsoft Copilot for individuals: your activity history, Copilot in Microsoft 365 apps for home: your data and privacy, Privacy FAQ for Microsoft Copilot (older app version), Data, Privacy, and Security for Microsoft Copilot (Microsoft..., Data protection when using Microsoft Copilot Chat for work or... · checked 7 Oct 2026
Perplexity
  • FreeYes, opt-out"AI data retention" is on by default; turn it off in Account settings > Preferences. Opt-out only covers data after that date.
  • PaidYes, opt-outSame for Pro and Max: AI data retention toggle in Preferences. Memory content may also be used to improve models.
  • TeamNoEnterprise data, including memory content, is never used for AI training.
Data Collection at Perplexity, Memory, Memory for Enterprise Organizations · checked 7 Oct 2026
Export and deleteCan you download your data and wipe what it remembers?What this means: Export is your exit: without it, switching assistants means starting from zero. Delete controls decide whether you can remove one memory or only everything.ChatGPT
  • FreeYesExport your history and data from Data Controls. Review, change or remove memories; delete chats or the account (removed within 30 days).
  • PaidYesExport your history and data from Data Controls. Review, change or remove memories; delete chats or the account (removed within 30 days).
  • TeamAdmin controlsAdmins can view, export and delete conversations in the workspace. Users can correct memory or delete a referenced chat.
OpenAI Privacy Policy, Enterprise privacy at OpenAI, Memory in ChatGPT, ChatGPT Enterprise and Edu release notes · checked 7 Oct 2026
Claude
  • FreeYesExport data in Settings > Privacy (web/desktop). Delete single memory topics, Reset memory, delete chats, or delete account.
  • PaidYesExport data in Settings > Privacy (web/desktop). Delete single memory topics, Reset memory, delete chats, or delete account. Memory data is included in data exports.
  • TeamAdmin controlsOnly the Primary Owner can export org data; members have no self-serve export. Members can delete their own memories and chats.
Export your Claude data, Use Claude's chat search and memory to build on previous context, Delete or rename a conversation, Export your organization's data · checked 7 Oct 2026
Gemini
  • FreeYesExport via Google Takeout (Gemini + My Activity > Gemini Apps). Delete any or all chats in Gemini Apps Activity; remembered facts go when their chats are deleted.
  • PaidYesExport via Google Takeout (Gemini + My Activity > Gemini Apps). Delete any or all chats in Gemini Apps Activity; remembered facts go when their chats are deleted. Also edit or delete each Instruction for Gemini in Settings > Personal Intelligence.
  • TeamAdmin controlsAdmin setting 'Allow users to delete conversations' (default on without Vault). Chat export for work accounts not documented.
Download your Gemini Apps data, Manage & delete your activity in Gemini Apps, Get personalization with memory of your past Gemini chats, Manage what Gemini remembers about you and customize responses, Control temporary chats and chat deletion in the Gemini App, Upload & analyze files in Gemini Apps · checked 7 Oct 2026
Copilot
  • FreeYesPrivacy dashboard: Export all activity history (CSV) or Delete all activity history. Delete single chats and memories in app.
  • PaidYesSame dashboard, plus export of Copilot in Microsoft 365 apps and Copilot Chat activity history.
  • TeamYesUsers delete activity history in My Account; admins search, export and delete memory via eDiscovery and Graph.
Manage your Copilot activity history in the privacy dashboard, Microsoft Copilot for individuals: your privacy controls and..., Delete your Microsoft Copilot activity history, Manage Copilot personalization and memory (Microsoft Learn) · checked 7 Oct 2026
Perplexity
  • FreeDelete onlyDelete single memories or clear all, delete threads or all threads, delete account. No self-serve data export (access via support).
  • PaidDelete onlyDelete single memories or clear all, delete threads or all threads, delete account. No self-serve data export (access via support). Individual threads can be exported to PDF for sharing; no full-account or memory export documented.
  • TeamAdmin controlsUsers can delete individual or all memories; admins can force deletion and set retention. Org export not documented.
Memory, Self-Serve Data Deletion, Account Deletion, GDPR Compliance at Perplexity, Security and Privacy with File Uploads, Memory for Enterprise Organizations, Data Collection at Perplexity · checked 7 Oct 2026
Temporary chatIs there a mode that skips memory and history?What this means: Use it for one-off or sensitive questions. It is not the same as private: most vendors still keep a copy for a short time.ChatGPT
  • FreeYesTemporary Chat: not in history, creates no memories, not used for training, deleted within 30 days. OpenAI pages differ on whether it can read existing memories.
  • PaidYesTemporary Chat: not in history, creates no memories, not used for training, deleted within 30 days. OpenAI pages differ on whether it can read existing memories.
  • TeamNot documentedThe business privacy page covers "unsaved conversations" (removed within 30 days) but no page we could read confirms Temporary Chat on business plans.
How your data is used to improve model performance, OpenAI Privacy Policy, Temporary chat in ChatGPT · checked 7 Oct 2026
Claude
  • FreeIncognitoIncognito chat (ghost icon): not saved to history or memory, not used for training, kept 30 days. Also a per-chat Memory off switch.
  • PaidIncognitoIncognito chat (ghost icon): not saved to history or memory, not used for training, kept 30 days. Also a per-chat Memory off switch. Not available inside projects; use the per-chat Memory toggle there instead.
  • TeamIncognitoAvailable, but included in org data exports and the Compliance API; kept 30 days or longer under org retention policy.
Use incognito chats, Use Claude's chat search and memory to build on previous context · checked 7 Oct 2026
Gemini
  • FreeYes'Temporary chat': not in Activity or recent chats, no personalization or memory, not used for training. Kept 72 hours.
  • PaidYes'Temporary chat': not in Activity or recent chats, no personalization or memory, not used for training. Kept 72 hours. Gems and Connected Apps can't be used in a temporary chat.
  • TeamAdmin controlsAdmins can allow temporary chats; with Vault, they are logged per Vault rules. User help still says not available for work accounts.
Use Gemini Apps (Start a temporary chat), Gemini Apps Privacy Hub, Control temporary chats and chat deletion in the Gemini App · checked 7 Oct 2026
Copilot
  • FreeYesTemporary chat skips memory and personalization and is not shown in chat history (shared Copilot help article). Retention period not stated.
  • PaidYesTemporary chat skips memory and personalization and is not shown in chat history (shared Copilot help article). Retention period not stated.
  • TeamYesTemporary chats may be retained under org retention policy and are discoverable by admins in Purview.
Personalize what Microsoft Copilot remembers, Manage Copilot personalization and memory (Microsoft Learn) · checked 7 Oct 2026
Perplexity
  • FreeIncognitoIncognito mode: memory and search history off, not used in memory, threads expire within 24 hours. Training use not stated.
  • PaidIncognitoSame Incognito mode for Pro and Max.
  • TeamIncognitoIncognito searches are never stored; admins can enforce Incognito for all searches by disabling search history.
Memory, Incognito Mode Troubleshooting, Memory for Enterprise Organizations, Data Retention and Privacy for Enterprise Organizations and U... · checked 7 Oct 2026
Compare two

How do two assistants compare on memory?

Two assistants can look alike and still differ on most rows: on paid plans, ChatGPT and Claude differ on 7 of the 10, including context window, file and project limits, and retention. Pick any two assistants and a plan. Rows where the short answers differ are marked, and you can hide the rows where they match.

Context windows

How big is each assistant's context window?

On free plans it runs from 27K tokens (ChatGPT, Instant model) to up to 1M (Claude); on paid plans Claude and Gemini Pro reach 1M, and Copilot and Perplexity publish no number at all. The context window is how much text one chat can hold at once, counted in tokens. It covers your messages, the replies, uploaded files and, on some apps, your memories and instructions, so the space left for your own text is smaller than the headline number. When a chat outgrows it, the oldest parts drop out of view, which is why long chats "forget" their beginning. More on that in what a context window is.

Context window in tokens by assistant and planLog scale. Bars show the context window each vendor documents for its chat app; grey means the vendor does not state a number for that plan.8K32K128K400K1MChatGPTFreeChatGPT Free: 27K (Instant) tokens, about 38 pages27K (Instant) ~38 pagesPaidChatGPT Paid: up to 400K (Pro) tokens, about 560 pagesup to 400K (Pro) ~560 pagesTeamNot documentedClaudeFreeClaude Free: up to 1M tokens, about 1,400 pagesup to 1M ~1,400 pagesPaidClaude Paid: 200K to 1M tokens, about 1,400 pages200K to 1M ~1,400 pagesTeamClaude Team: 200K to 1M tokens, about 1,400 pages200K to 1M ~1,400 pagesGeminiFreeGemini Free: 32K tokens, about 45 pages32K ~45 pagesPaidGemini Paid: 128K to 1M tokens, about 1,400 pages128K to 1M ~1,400 pagesTeamGemini Team: 32K to 1M tokens, about 1,400 pages32K to 1M ~1,400 pagesCopilotFreeNot documentedPaidNot documentedTeamNot documentedPerplexityFreeNot documentedPaidNot documentedTeamNot documented

Log scale. Pages assume about 70 words per 100 tokens (Google's rule of thumb in its token docs) and 500 words per page, so treat them as rough. Where a plan lists several numbers, the bar shows the largest and the label says which plan gets it.

What it means for you

When should this table change what you do?

In three situations: when you switch assistants, when you paste sensitive work into a chat, and when you choose a plan for a team. Each one comes down to two or three rows.

You are switching assistants

Memories do not travel on their own. Check the Export and delete row first: ChatGPT, Claude, Gemini and Copilot let you export; Perplexity documents no self-serve export. Gemini can import memories and chat history from other AI apps. If you would rather keep one memory that no single vendor owns, a local app such as Locul keeps it as files on your own machine and serves it to Claude, ChatGPT and Cursor over MCP. Our memory transfer guide covers what carries over.

You paste sensitive work into chats

Read three rows together: Training, Retention and Temporary chat. On personal plans, four of the five may train on your chats unless you opt out, so the setting matters more than the plan. A temporary chat keeps things out of memory, but it is still stored for a while, from 24 hours to 30 days.

You are choosing a plan for a team

Switch the table to Team. Memory is admin-controlled almost everywhere, and Gemini on Workspace has no memory, past-chat recall or personal instructions at all. Claude's memory has to be enabled by the org owner and then switched on by each member, and Perplexity's enterprise memories belong to the organization, not the person.

Method

How was every cell checked?

Every cell was checked by hand against the vendor's own documentation: we read 88 vendor pages on 7 October 2026 and saved 249 verbatim quotes behind the 150 cells. The pages are each vendor's current help center plan pages, privacy pages and release notes for the five consumer chat apps (ChatGPT, Claude, Gemini, Microsoft Copilot and Perplexity) and their business versions. For every cell we saved the exact page and the sentence that supports it. Those quotes ship with the data in data.json, so anyone can check a cell without trusting us.

The rules we held to

  • Vendor pages only. No third-party blogs, no forum posts, no answers from the assistants themselves. Two vendors block automated readers. Perplexity's help center on perplexity.ai does, so we read the same articles on its official Intercom-hosted help center and link there. OpenAI's help center and plan pages do too, so for ChatGPT we read OpenAI's plan, privacy and policy pages as captured by the Internet Archive between 1 and 5 October 2026 (each snapshot is listed in the data file), plus help-center text as shown in search results, marked as such. Cells link to the live vendor page. We never got past a bot check to read anything.
  • "Not documented" means not documented. If no vendor page we could find states a number or a default, the cell says so. We did not fill gaps from memory, from the API docs, or from testing a single account, because those vary by account, region and rollout.
  • The chat app, not the API. Context windows are the ones the vendor gives for its chat app on each plan. API limits are often larger and are not what you get in the app.
  • Three plan groups. Free is the signed-in free tier. Paid groups the individual plans (for example Plus and Pro, or Pro and Max); where they differ, the cell names the plan. Team groups the business and enterprise plans, where an admin often decides what is on.
  • Plain English. Cells paraphrase the vendor wording into one line. Vendor quotes are kept verbatim in the data file, with long dashes swapped for hyphens.

What this table does not tell you

Features roll out in waves, so a feature listed here may not have reached your account or your country yet. Gemini's saved memories, for example, are documented as US only. Admins on business plans can switch most of these features on or off for everyone. And a documented limit is a ceiling, not a promise: long chats can still lose detail well before they reach the context window.

Changelog

What changed, and when was this last checked?

Last checked on 7 October 2026. This first edition logs the vendor changes from the past few months that shaped it. The table is re-checked against every source at least once a month and whenever a vendor announces a memory or plan change. Each refresh adds an entry here, so you can see what moved.

  1. First edition. All 150 cells (5 assistants, 10 features, 3 plan groups) checked against the vendors' own pages.

    Claude, 25 Aug 2026: one memory across chat and Cowork, editable memory topics, sensitive topics left out unless you opt in, and memory off by default for Team and Enterprise members. Source

    Claude, 2 Mar 2026: memory from chat history opened to all users, including Free. Source

    Gemini, 30 Sep 2026: skills rolled out in Gemini chat. Gems move to skills from November 2026 on personal accounts, March 2027 on Workspace business and June 2027 on education. Source

    Gemini, 26 Mar 2026: you can import memories and chat history from other AI apps into Gemini. Source

    Gemini: a Google AI Plus tier with a 128K token context window now sits between the free app (32K) and Pro or Ultra (1M). Source

    Copilot, 18 Aug 2026: a new Microsoft Copilot app with its own privacy pages. Microsoft says prompts, responses and files in the new app are not used to train foundation models. Source

    Copilot: Microsoft 365 Copilot is now called Microsoft Copilot, and Microsoft 365 Copilot Chat is now Microsoft Copilot Chat. Source

    Perplexity, 18 Jun 2026: memory for Enterprise organizations, with admin controls. Memories created in an organization belong to the organization, not the person. Source

Sources

Every vendor page behind the table

Only the vendors' own help centers, docs, plan pages and privacy pages. No third-party summaries.

ChatGPT 16 pages

Claude 20 pages

Gemini 16 pages

Copilot 18 pages

Perplexity 18 pages

FAQ

Questions people ask about AI memory

Does Gemini have better memory than ChatGPT?

Not on the free plans. ChatGPT Free keeps saved memories (marked "Limited") and a lightweight memory of recent chats. Gemini's free app remembers past chats, but saving explicit facts with "remember that..." needs a Google AI plan and works only in the US. On paid plans they are closer: both remember past chats and saved facts, and Gemini Pro and Ultra have a much larger context window (1M tokens against 54K to 400K on ChatGPT's paid plans, depending on plan and model).

What is the difference between Claude's memory and ChatGPT's memory?

Both keep facts about you across chats and both let you view and delete them. Claude stores memory as topics you can read and edit in Settings, keeps a separate memory for each project, and since 2 March 2026 does so on the Free plan too. Searching past chats is on paid Claude plans only. ChatGPT has two switches, saved memories and chat history, and gives Free users a lighter version of both. See ChatGPT memory vs Claude memory for how each behaves day to day.

Does Claude have memory on the free plan?

Yes. Anthropic's help center says memory is on by default for Free, Pro and Max on the web, desktop and mobile apps. What Free does not get is searching and referencing past chats, which is documented for paid plans only. On Team and Enterprise an owner has to enable memory first, and then each member turns it on. More in does Claude have memory.

How big is ChatGPT's context window?

OpenAI's plan page gives totals per plan: 27K tokens on Free with the Instant model, 54K on Go and Plus (256K with the reasoning model), and 128K on Pro (400K with reasoning). Those totals include system instructions, memories and the model's reasoning, so the space for your own text is smaller. OpenAI does not state a number for Business and Enterprise on the pages we could read.

Does Perplexity remember previous conversations?

Yes. Perplexity's Memory saves details between conversations and can also reference your search history, each with its own switch under Personalize. You can delete single memories or clear them all. In Incognito mode both are off and threads expire within 24 hours. Perplexity does not publish a context window for any plan.

Does Microsoft Copilot remember what I tell it?

Yes. Copilot Memory is on by default in the Copilot app, and Copilot can also infer things from your chat history. You can delete one memory or all of them in Settings. On work accounts, Microsoft says deleting a chat does not delete a saved memory, and admins can switch memory off for everyone.

Are my chats used to train AI models?

On personal plans, ChatGPT, Claude, Gemini and Perplexity may use your chats unless you opt out, each with its own setting: "Improve the model for everyone" in ChatGPT, "Help improve our AI models" in Claude, Keep Activity in Gemini and "AI data retention" in Perplexity. Microsoft says prompts in the new Copilot app are not used to train foundation models. On business plans, all five say they do not train on your data by default.

How do I stop an AI assistant from remembering something?

Three ways, from lightest to strongest: use the temporary or incognito chat mode for that conversation, delete the specific memory in settings, or turn memory and past-chat recall off. Note that Gemini ties memories to chats, so you delete the chats that contain the detail, while Claude and Copilot keep memories even after the chat that created them is deleted.

Why does my AI forget things in a long chat?

Because of the context window. Each chat can only hold so many tokens, and once a conversation, its files and its instructions go past that size, the oldest parts fall out of what the model can see. Memory features help across chats, but they store short summaries, not the whole conversation. Starting a fresh chat with a short summary, or putting key files in a project, usually works better than one endless thread.

Use the data

Download, embed or cite this table

You can reuse the data under CC BY 4.0. Link back to this page so readers get the latest check.

Cite it

Khalid, J. (2026). AI memory and context limits: ChatGPT, Claude, Gemini, Copilot and Perplexity, checked against vendor documentation on 7 Oct 2026. Locul. https://locul.ai/research/ai-memory-limits/

Download

Spotted a change? Email [email protected] with the vendor link and it goes into the next refresh.

Related reading