# ChatGPT Deep Research Limits: Plan Quotas, Reset Rules, and Workspace Solutions

ChatGPT Deep Research limits restrict autonomous research tasks through plan-based quotas, rolling 30-day reset cycles, and document ingestion boundaries. Personal plans provide monthly task allocations that meter multi-step queries separately from everyday conversation limits, while business tiers draw from shared credits. Teams researching large document archives can bypass file attachment ceilings by connecting external repositories via Model Context Protocol workspaces.

Source: https://fast.io/resources/chatgpt-deep-research-limits/
Author: [Tom Langridge](https://fast.io/authors/tom-langridge/)
Last reviewed: 2026-09-22

## How Do ChatGPT Deep Research Limits and Rolling Resets Work?

ChatGPT Deep Research limits are the plan-dependent task caps, rolling 30-day reset cycles, and source-fetching quotas enforced by OpenAI on its autonomous multi-step research agent. Unlike standard conversational prompts that draw from short-term messaging windows, Deep Research runs on a dedicated metering system because a single query triggers dozens of search operations, site browses, and document syntheses that can run for 5 to 30 minutes. OpenAI separates these execution allowances from everyday chat limits to prevent autonomous reasoning chains from monopolizing inference clusters.

OpenAI's own documentation states that the in-product usage counter shows your remaining tasks, and that on plans with a fixed monthly allowance it resets every 30 days from the date of your first use. Understanding how this rolling counter behaves is necessary for teams planning intensive research sprints, competitor sweeps, or technical due diligence.

### The Mechanics of the Rolling 30-Day Reset Cycle

A common misconception among subscribers is assuming that Deep Research allowances reset on the first day of the calendar month or synchronize with their credit card billing cycle. In practice, the 30-day reset window is personal, rolling, and tied to the timestamp of your first research query.

When your allocation unlocks, the timer remains idle until you dispatch your first Deep Research task. Executing a task on September 15 locks that specific query credit into a rolling 30-day window that replenishes on October 15. If you consume your entire monthly allocation within the first 48 hours of a project sprint, your account remains throttled for the remainder of that 30-day window.

This rolling schedule prevents users from hoarding queries or timing large batch runs around billing dates. To keep track of your schedule, monitor the in-product counter before launching multi-step queries:

* **In-Product Counter Location:** Navigate to the model picker or account settings panel inside ChatGPT. The interface displays your remaining active tasks and indicates the date when the next query slot replenishes.
* **Non-Cumulative Allowances:** Unused research queries do not roll over into subsequent months. If your tier provides a monthly allocation and you execute zero tasks, your counter does not expand during the next cycle.
* **Independent Quota Pools:** Running out of standard conversational messages for models like GPT-4o or GPT-5 does not block Deep Research. Similarly, burning your entire research allowance leaves your regular chat, coding, and image generation capabilities fully functional.

### Full Research Tasks vs Lightweight Model Fallbacks

OpenAI applies different throttling behaviors depending on your active plan tier. On personal tiers, consuming your full research allowance typically pauses autonomous web synthesis until the counter resets. On certain plans, the system provides access to a lightweight fallback mode once your primary allocation is depleted.

Lightweight research tasks rely on smaller, faster reasoning models that synthesize queries using fewer search steps and shallower browsing depth. While lightweight tasks return structured reports faster, they lack the multi-layered source cross-referencing and extensive synthesis capabilities of full research runs. For corporate decision-makers evaluating market opportunities or legal obligations, relying on lightweight fallbacks can result in missed citations and incomplete analysis.

### Tracking Query Depletion During Active Sprints

Because Deep Research queries operate autonomously in the background, users can easily lose track of their monthly consumption during rapid investigation cycles. Each distinct prompt submitted with the Deep Research tool active counts as a full task deduction, regardless of whether the agent runs for five minutes or thirty minutes.

If an initial prompt is poorly specified, cancelling the run mid-flight or immediately refining the query consumes a second task from your quota. To protect your monthly allowance, spend time refining research prompts before submission, specifying target domains, required data points, and explicit formatting expectations.

## What Are the Deep Research Limits Across Free, Plus, Pro, and Business?

OpenAI structures Deep Research access across subscription tiers to balance compute expenditure against user demand. Rather than publishing a task count per plan, its pricing page now uses descriptive tier definitions: Free and Go get "Limited deep research", Plus gets "Expanded deep research", and Pro gets "Maximum deep research". The in-product counter, not the pricing page, is where you read your actual remaining tasks.

The table below outlines how Deep Research limits, metering rules, and ingestion constraints operate across all official ChatGPT plan tiers:

| Plan Tier | Published Access Level | Metering & Reset Cycle | Context & File Scope | Status (September 2026) |
| :--- | :--- | :--- | :--- | :--- |
| **Free ($0) & Go ($8/mo)** | Limited access | Rolling 30-day counter | Standard file upload caps | Active evaluation tier |
| **Plus ($20/mo)** | Expanded access | Rolling 30-day counter from first query | Standard file upload caps | Active individual plan |
| **Pro (from $100/mo)** | Maximum access | Rolling 30-day counter from first query | Standard file upload caps | Active primary Pro plan |
| **Business Standard** | Included baseline access | Billed at 50 workspace credits per extra task | Connected apps (Drive, SharePoint) or MCP | Active team plan ($20 to $25/seat) |
| **Edu** | Campus allocation | Rolling 24-hour window (5 queries per 24 hours) | Campus project scope | Active academic tier |

### Individual Plans: Free, Go, Plus, and Pro

Individual accounts receive fixed monthly query allocations that meter autonomous research tasks in the main chat interface:

* **Free and Go plans:** These entry tiers offer strictly limited evaluation access for basic inquiries. Users can test autonomous research on narrow questions, but the system quickly transitions to lightweight models or pauses access until the rolling 30-day reset arrives.
* **Plus plan:** Designed for individual professionals, the Plus plan provides an expanded task allowance. This tier suits occasional market scans, vendor evaluations, and technical research, provided users space their queries across the month.
* **Pro plan:** Pro plan subscribers receive the maximum research allocation available for individuals. In September 2026, OpenAI paused new sign-ups and upgrades for the higher Pro tier, leaving the standard Pro subscription as the primary active option for professionals seeking top-tier compute. Pro accounts provide priority execution during peak demand windows and deeper browsing depth.

### Team and Enterprise Tiers: Business Credits and Edu Quotas

For organizational teams, Deep Research pairs with shared workspace billing and centralized administration:

* **ChatGPT Business plan:** Business Standard seats include a baseline allocation of Deep Research queries for each team member. When an organization requires additional capacity, tasks draw from shared workspace credits. OpenAI's published rate card charges one deep research task at approximately 50 credits. Business workspace credits are purchased and managed under workspace billing settings, providing flexible burst capacity for major research initiatives.
* **Edu plan accounts:** Academic accounts follow a distinct metering model designed for university campuses. Rather than a 30-day cycle, Edu plans provide 5 deep research queries per rolling 24 hours at no credit cost, allowing students and faculty to conduct regular academic literature reviews without monthly quota anxiety.

### Deep Research in Personal Chat vs ChatGPT Work

A critical operational distinction exists between running Deep Research in standard personal Chat versus inside ChatGPT Work or Codex environments. In standard Chat, queries draw directly from your personal plan allowance.

In ChatGPT Work, Deep Research draws from your shared Work and Codex allowance or organizational credits instead of touching your personal Chat quota. This dual-metering architecture allows team members who have exhausted their personal monthly tasks to continue executing project research inside shared organizational workspaces.

## What Are the File Upload and Document Ingestion Limits in Deep Research?

While Deep Research is primarily known for browsing the public web, many high-value research tasks require analyzing private internal files, such as financial statements, product specifications, or customer contracts. However, feeding private documents into ChatGPT exposes users to strict file upload limits and context window constraints.

The table below summarizes the technical upload limits and document ingestion boundaries enforced across ChatGPT:

| Ingestion Boundary | File Constraint | Impact on Multi-Step Research |
| :--- | :--- | :--- |
| **Individual Document Size** | 512 MB per file | Large PDF archives must be split before upload |
| **Document Token Density** | 2 million tokens per document | Text truncation occurs when reading long manuscripts |
| **Tabular Spreadsheets** | Approximately 50 MB | Formula parsing memory overhead halts processing |
| **Reference Images** | 20 MB per image | Resolution downsampling limits diagram analysis |
| **Platform Ingestion Velocity** | 80 files per 3 hours | Rapid bulk ingestion triggers temporary rate limits |
| **Project Knowledge Base** | 5 files (Free), 25 (Go, Plus), 40 (Edu, Pro, Business, Enterprise) | Comprehensive document archives exceed project caps |

### Understanding File Size and Ingestion Bottlenecks

Beyond individual file size constraints, ChatGPT limits how many files can reside within a single project knowledge base. OpenAI publishes 5 files per project on Free, 25 on Go and Plus, and 40 on Edu, Pro, Business, and Enterprise, with only 10 files uploadable at a time.

These document count ceilings create immediate bottlenecks for data-intensive research:

* **Individual Documents:** File sizes and token densities are restricted to avoid saturating context buffers during automated analysis.
* **Complex Spreadsheets:** Tabular data faces strict memory constraints during multi-turn parsing and calculation.
* **Image Exhibits:** Visual attachments are evaluated through multimodal extraction models subject to resolution ceilings.
* **Platform Ingestion Velocity:** Upload frequency is metered over rolling multi-hour windows to manage server bandwidth.
* **Project File Ceilings:** Project knowledge bases permit expanded document counts on higher plans, but multi-hundred-document archives require external indexing.

An enterprise compliance review, a historical patent analysis, or an acquisition audit routinely involves hundreds of PDF reports and spreadsheets. Attempting to upload an entire document repository into a ChatGPT Project triggers upload rejections once you hit platform thresholds.

### Connected App Permissions and Security Boundaries

To help teams work around manual upload caps, OpenAI allows ChatGPT Deep Research to connect directly to cloud storage platforms, including Google Drive, Microsoft OneDrive, and Microsoft SharePoint. When connected apps are enabled by workspace administrators, Deep Research can inspect files stored in corporate drives alongside its public web searches.

However, connected apps operate under strict operational boundaries:

* **Read-Only Operations:** Deep Research uses read actions exclusively. The agent can search, retrieve, and analyze cloud documents, but it cannot write new files, update folders, or alter metadata in your cloud storage.
* **Identity Scoping:** The agent only accesses documents that the individual authenticated user has explicit permission to read in the underlying cloud service.
* **Mandatory Web Search:** OpenAI admin specifications require web search to remain active for Deep Research to function, even when the query focuses entirely on internal connected files.
* **API Latency and Timeout Vulnerabilities:** Large file traversals across third-party cloud APIs often hit platform request timeouts, causing multi-step research runs to stall mid-execution.

### Restricting Research Scopes to Trusted Domains

To prevent Deep Research from wasting source quotas on low-quality websites, OpenAI introduced source-filtering controls. Users can navigate to Sites and select Manage sites to explicitly restrict the agent's web search to verified domains.

Restricting searches to authoritative domains, such as regulatory portals, academic repositories, or verified corporate newsrooms, reduces hallucinations and ensures that research tokens are spent analyzing primary evidence rather than aggregator summaries.

## Why Do Research Tasks Burn Quota on Document Repositories?

When organizations attempt to use ChatGPT Deep Research for internal document analysis, they frequently exhaust their monthly task allocations while achieving disappointing results. Understanding the technical reasons behind this inefficiency reveals why autonomous web-browsing agents struggle when pointed at large unstructured file collections.

Deep Research was engineered primarily for broad web discovery rather than dense private repository retrieval. When an agent attempts to analyze dozens of uploaded files, several compounding factors rapidly drain token budgets and task allowances.

### Compounding Transcript Bloat in Agentic Reasoning Loops

Deep Research does not process prompts in a single forward pass. Instead, the underlying reasoning model formulates a plan, executes searches, inspects files, evaluates findings, and iterates. In an autonomous trajectory lasting twenty or thirty steps, token consumption compounds exponentially.

Every file snippet or search result retrieved by the agent is appended to the active conversational transcript. On subsequent steps, the entire accumulated transcript must be fed back into the model as input tokens. If an agent inspects ten extensive PDF documents during its run, the active prompt payload swells by tens of thousands of tokens on every single reasoning step.

This compounding token expansion creates two severe problems:

1. **Context Window Thrashing:** As the transcript approaches the model's context ceiling, the system applies lossy summarization to compress earlier conversation steps. During this compression, specific numbers, contractual exceptions, and subtle nuances get discarded, leading to hallucinations or incorrect inferences in the final report.
2. **Quota Waste on Trivial Reads:** An entire Deep Research task allowance is consumed simply to locate and extract three relevant sentences from a lengthy operational manual. Spending a dedicated monthly research task on internal document retrieval is an expensive misuse of compute.

### The Refinement and Re-Run Penalty

In real-world business environments, research is rarely complete after a single prompt. Stakeholders frequently review initial findings and require follow-up analysis: comparing a newly discovered clause against earlier quarters, verifying pricing tiers, or exploring adjacent regulatory requirements.

Because Deep Research treats each prompt dispatch as a standalone task, every refinement prompt consumes another full query from your monthly allowance. Running three exploratory drafts and two refined analyses can wipe out an entire month's allocation of tasks on standard plans in a single afternoon. Teams need a decoupled architecture where document indexing happens once, allowing agents to query specific facts repeatedly without burning full research allocations.

### Source Verification Overhead on Unindexed Files

Autonomous research agents frequently struggle with confidence calibration, presenting inferred facts with the same authoritative tone as verified citations. When Deep Research parses raw uploaded documents without structured indexing, it can blend adjacent paragraphs or attribute findings to the wrong page.

Verifying claims in an unindexed twenty-page generated report forces human reviewers to manually cross-check every cited source against original PDFs, defeating the time-saving purpose of autonomous research.

## How Do You Connect Large File Collections Without Hitting ChatGPT Caps?

The most effective way to analyze large corporate file archives with AI agents is to decouple file storage and indexing from model inference. Instead of uploading dozens of PDFs into a ChatGPT Project or burning monthly Deep Research tasks to scan raw folders, organizations can store their document corpus in an intelligent workspace built for agent collaboration.

Intelligent workspaces index documents upon arrival, transforming static files into an active knowledge base that agents query on demand. By connecting external storage to agents through open standards, teams eliminate file-count caps, protect their research task allowances, and ground AI responses in verified citations. Learn more about purpose-built agent infrastructure on [Fast.io storage for agents](/storage-for-agents/).

### Managing Large Document Archives in Fast.io Workspaces

Fast.io provides an intelligent workspace platform designed for agentic teams and human collaborators. Rather than forcing users to manually manage file attachments across different AI chats, Fast.io serves as a centralized coordination layer for organizational knowledge:

* **Shared Organization-Owned Workspaces:** Workspaces hold large document collections, project archives, and client deliverables without artificial file-count caps or context window restrictions.
* **Automated URL and Cloud Import:** Organizations can import documents directly from Google Drive, Microsoft OneDrive, Box, or Dropbox via URL without local file transfers or storage overhead.
* **Intelligence Mode and Hybrid Search:** When Intelligence Mode is enabled on a workspace, files are indexed automatically. Fast.io combines full-text keyword matching, semantic meaning, and metadata value search. When an AI queries the workspace, the search engine returns exact text excerpts with source document citations.
* **Metadata Views for Structured Data:** For teams handling high volumes of contracts, financial forms, or technical specifications, Fast.io Metadata Views turn unstructured documents into live, queryable databases. Users describe required fields in natural language, and the system extracts structured values across PDFs, scans, and spreadsheets. Learn more about automated document extraction on [Fast.io Metadata Views](/product/document-data-extraction/).

### Connecting Agents via the Remote Model Context Protocol Server

Rather than uploading raw files into ChatGPT or Claude, agents connect directly to Fast.io workspaces through the Model Context Protocol (MCP). Fast.io hosts a remote MCP server accessible over Streamable HTTP at `https://mcp.fast.io/mcp` (or `https://mcp.fast.io/mcp/key` with bearer token authentication), alongside legacy SSE support at `https://mcp.fast.io/sse`.

When configured with Fast.io MCP tools, an agent executes targeted retrieval actions rather than ingesting entire file corpuses into its prompt. For example, an agent researching vendor compliance calls the `storage/search` tool, retrieves the exact three relevant clauses from a five-hundred-page contract repository, and answers the user with authoritative citations.

Here is an example configuration connecting an MCP-compatible agent environment to a Fast.io workspace using Streamable HTTP:

```json
{
  "mcpServers": {
    "fastio": {
      "url": "https://mcp.fast.io/mcp/key",
      "headers": {
        "Authorization": "Bearer YOUR_FASTIO_API_KEY"
      }
    }
  }
}
```

This decoupled architecture preserves ChatGPT Deep Research allowances for broad external web discovery while routing internal document queries through indexed MCP storage. Multi-agent workflows can write research deliverables, summaries, and audit records directly back into shared workspaces, where team members review outputs using per-file version history and an append-only audit log.

For command-line automation and pipeline scripting, developers can install the official `@vividengine/fastio-cli` package. Every organization starts with a 14-day free trial, which requires a credit card. Subscription tiers include Starter, Business, and Enterprise plans, with complete details on the [Fast.io pricing page](/pricing/).

### Step-by-Step Implementation for Agentic Research

Deploying an indexed workspace for AI research involves three practical steps:

1. **Create an Organization Workspace:** Sign up for an organization workspace and import your research corpus from local storage or cloud drives using one-time URL import.
2. **Enable Intelligence Mode:** Toggle Intelligence on the workspace to initiate background indexing across all uploaded PDFs, presentations, and technical documentation.
3. **Connect Your Agent via MCP:** Add the Fast.io remote MCP endpoint to your client configuration. Your agents can immediately search, retrieve, and cite verified excerpts across thousands of internal documents without uploading a single file to chat.

## Frequently asked questions

### How many Deep Research tasks do you get in ChatGPT per month?

OpenAI no longer publishes a single universal query count across all tiers. On personal plans, access is tiered: Free and Go accounts receive a limited evaluation allowance, Plus subscribers receive an expanded allocation, and Pro accounts receive maximum access. ChatGPT Business includes baseline research tasks per seat and meters additional runs at about 50 workspace credits per task. Edu accounts receive 5 queries per rolling 24 hours.

### When do ChatGPT Deep Research limits reset?

Deep Research limits operate on a rolling 30-day reset cycle rather than calendar months or billing dates. The 30-day window starts from the exact timestamp of your first research query within that period. Tasks replenish 30 days after they were used, meaning allowances restore continuously rather than resetting in a single batch on the 1st of the month.

### What happens when you exceed your Deep Research quota?

When your monthly Deep Research allowance is exhausted, execution is typically paused until your rolling 30-day window replenishes. On certain personal plans, queries may fall back to lightweight models powered by smaller reasoning engines. On ChatGPT Business, organizations can continue researching by drawing from shared workspace credits.

### Can ChatGPT Deep Research analyze private company files without uploading them to chat?

Yes, Deep Research can inspect files in supported connected apps such as Google Drive and Microsoft SharePoint, provided administrators enable them and the user holds read access. However, connected apps remain subject to API timeouts and require web search to remain active. A more reliable solution is indexing documents in a shared Fast.io workspace and querying them via Model Context Protocol.

### What is the difference between Deep Research in Chat and ChatGPT Work?

In standard Chat, Deep Research draws from your personal plan allowance. In ChatGPT Work or Codex, Deep Research draws from your shared Work and Codex allowance or organization credits instead of touching your personal Chat quota. This allows team members who have exhausted personal Chat limits to continue researching within team projects.

### How much does an extra Deep Research task cost on ChatGPT Business?

On ChatGPT Business, extra Deep Research queries beyond the included seat allowance draw from workspace credits. OpenAI's rate card lists a deep research task at about 50 workspace credits. Credits are purchased and managed under workspace billing settings.

## Sources

- [OpenAI Help Center: Deep research in ChatGPT](https://help.openai.com/en/articles/10500283-deep-research-in-chatgpt) — OpenAI shows an in-product counter for remaining Deep Research tasks, and on plans with a fixed monthly allowance it resets 30 days after the user's first use.
- [OpenAI Help Center: ChatGPT Rate Card (Business, Enterprise/Edu credit-based pricing)](https://help.openai.com/en/articles/11481834-chatgpt-rate-card-business-enterpriseedu-credit-based-pricing) — On credit-based Business and Enterprise plans, OpenAI's rate card charges one Deep Research task at approximately 50 credits.
- [OpenAI Help Center: ChatGPT Rate Card (Business, Enterprise/Edu credit-based pricing)](https://help.openai.com/en/articles/11481834-chatgpt-rate-card-business-enterpriseedu-credit-based-pricing) — ChatGPT Edu customers receive 5 Deep Research queries per rolling 24-hour period at no credit cost.
- [OpenAI Help Center: Projects in ChatGPT](https://help.openai.com/en/articles/10169521-projects-in-chatgpt) — ChatGPT Projects cap uploads at 5 files per project on Free, 25 on Go and Plus, and 40 on Edu, Pro, Business and Enterprise.
- [OpenAI: ChatGPT Pricing](https://openai.com/chatgpt/pricing/) — OpenAI describes Deep Research access as Limited on Free and Go, Expanded on Plus, and Maximum on Pro, with Pro priced from $100 per month.

## About Fast.io

Fast.io provides shared workspaces where people and AI agents work on the same files, with built-in semantic search and citation-backed chat over what they hold. Agents reach it through a remote MCP server at https://mcp.fast.io/mcp, a REST API at https://api.fast.io/current/, and a command line client published on npm as @vividengine/fastio-cli.