How many Deep Research sessions can you run on ChatGPT?

The subscription you choose dictates two distinct buckets of usage, with a fallback to a lighter engine once the main allowance is gone.Official documentation updated last month confirms that Free users receive five tasks every thirty days, all handled by a lightweight o4‑mini model. 

Plus, Team, Enterprise, and Education members begin each cycle with ten full‑model tasks powered by o3; after those are consumed, ChatGPT automatically grants fifteen additional lightweight tasks. Pro subscribers start much higher—one‑hundred‑twenty‑five full‑model tasks—and, once those expire, they gain another one‑hundred‑twenty‑five lightweight runs.

Plan	Full‑model quota (o3)	Lightweight quota (o4‑mini)	Total tasks per 30‑day window
Free	—	5	5
Plus / Team / Enterprise / Edu	10	15	25
Pro	125	125	250

The lightweight mode produces shorter, less resource‑intensive reports but retains citations and multi‑source synthesis. Industry trackers and community posts echo these figures, noting the jump from the original ten to a combined twenty‑five tasks for paid non‑Pro tiers earlier this spring.

The rolling thirty‑day window begins with your first request

Once you trigger Deep Research, an invisible clock starts; exactly thirty days later every counter resets to zero.Unlike the weekly caps applied to ordinary o‑series chat, Deep Research works on a personal, rolling calendar: initiate your first study on 3 August at 09:00 UTC, and all credits refresh automatically at 09:00 UTC on 2 September. Switching tiers mid‑cycle does not restart that clock; the remaining balance simply converts to the new plan’s limits.

Automatic fallback keeps the tool available after the main quota expires

When your full‑model allowance reaches zero, subsequent clicks silently downgrade to the lightweight engine until the cycle renews.This hand‑off means Plus‑level users never see an outright block; they just receive shorter, faster reports. Pro users experience the same transition near the very end of an intensive month—rare in practice given the high ceiling of two‑hundred‑fifty total tasks. A banner appears only when both buckets are empty, stating when new sessions will be available.

One credit equals one autonomous investigation, regardless of duration

Whether your study finishes in five minutes or half an hour, it still counts as a single task.Attaching large PDFs, spreadsheets, or images does not multiply the charge. Interactive follow‑up questions, however, launch new Deep Research runs if you press the tool button again—so grouping sub‑questions into one well‑scoped brief is the most efficient strategy.

Careful prompt design stretches every allowance

Thoughtful scoping and timing can make ten full tasks feel like twenty.

• Draft a single, comprehensive brief instead of a string of narrow prompts; the agent can branch internally.

• Upload supporting documents to reduce web‑crawl time and keep the agent focused.

• Use GPT‑4o with web search for quick facts, reserving Deep Research for analyses that truly need multistep verification.

• Schedule large studies soon after a reset so you’re never caught mid‑project without credits.

Seasoned users on community forums report that batching questions and leveraging lightweight mode for preliminary scans preserves the full‑model quota for mission‑critical reports.

__________

FOLLOW US FOR MORE.

DATA STUDIOS

datastudios.org