AI Deep Research Comparison — ChatGPT vs Gemini vs Perplexity vs Claude vs Grok

“Deep research” is the agentic mode in a chatbot that spends several minutes reading tens of web pages and then writes a long, structured, cited report — a different job from quick AI search, which answers in seconds from a handful of pages. This page lines up the five modes a Sri Lankan student, analyst or freelancer actually chooses between: ChatGPT Deep Research, Gemini Deep Research, Perplexity Deep Research, Claude Research and Grok DeepSearch.

1. The dataset

The comparison is a curated, cited table — not a benchmark and not an opinion score. Each row records the underlying model, the cheapest plan that unlocks deep research, the free-tier allowance, the paid run limit, typical report length, citation style, export targets, private-data connectors and real-time-web access, taken straight from the vendor's own pages. Because deep-research limits change often, every figure is dated with a 2026-07-05 LAST_VERIFIED stamp.

2. The recommender

The tool is a transparent, rules-based recommender — no live API calls, no scoring model. It runs in two steps:

1. Filter.For each tool it finds the cheapest plan whose price is within your budget, whose monthly run limit meets your minimum, and — if you asked — is free. Tools without document export or private connectors are dropped when you require those. A tool with no qualifying plan is set aside as a “near miss” with the exact constraint it failed.
2. Rank.Survivors are sorted by a key order tied to your use case — Academic report leads on report depth then export then citation quality; Market research leads on runs per month then sources per run; News & real-time leads on live-web freshness. Ties break by lower price, then more runs.

The top survivor becomes the recommendation and the second the runner-up. If nothing clears every filter, the tool shows the nearest options and which constraint each fails, so the page never dead-ends.

3. The LKR line is indicative

The recommendation card multiplies the USD plan price by a fixed CBSL middle rate of Rs 300per USD to show an approximate rupee cost. This is a constant, not a live feed — your bank's rate plus FX margin will differ by a few percent. For a live USD-to-LKR rate with Wise, Payoneer and Skrill fees, use the Freelancer USD-LKR calculator in Related tools.

4. Cross-checks

The data module exports verifyWorkedExamples() — which recomputes the worked examples below, including the fractional budget, negative-input error, runs-boundary and no-match cases — and verifyDatasetIntegrity(), which asserts unique ids, non-negative prices and runs, axis scores in range, and an https source on every row. A typo in a quarterly price update fails a test instead of shipping silently.

• OpenAI — ChatGPT Deep Research FAQ (run limits by plan)
• OpenAI — ChatGPT pricing (Free, Plus, Pro)
• Google — Gemini Deep Research overview & Docs export
• Google — AI plans (AI Pro, AI Ultra)
• Perplexity — Introducing Perplexity Deep Research
• Perplexity — Help Center & FAQ (free vs Pro limits)
• Anthropic — Claude Research
• Anthropic — Claude pricing (Pro, Max)
• xAI — Grok plans & DeepSearch (Free, SuperGrok)
• Columbia Journalism Review (Tow Center) — AI citation-accuracy study
• Central Bank of Sri Lanka — Exchange Rates (indicative LKR line)

Plans, run limits, citation behaviour and export options were last cross-checked against the official sources on 2026-07-05. Consumer tiers only; Team, Enterprise and API plans are out of scope. LKR is indicative at Rs 300/USD.