OpenAI Killed Legacy Deep Research — Here's What to Use Instead

Legacy deep research dies March 26. Current version stays, old chats safe. How ChatGPT's 25 queries/mo compares to Claude, Gemini, and Perplexity.

If you open ChatGPT tomorrow and the Deep Research button looks different — or gone entirely — don’t panic. OpenAI is removing the legacy deep research mode on March 26, 2026. The current deep research still works. Your old conversations and reports are safe. But the version that launched with o3 back in early 2025 is being retired.

Here’s what’s actually changing, what stays the same, and what alternatives exist if you want to compare options.

What’s Being Removed

OpenAI is deprecating the legacy deep research mode — the original version that launched in February 2025 as a specialized o3-based model. This was the first iteration that could autonomously browse the web for 5-30 minutes, scan hundreds of sources, and generate cited research reports.

What’s NOT being removed: the current deep research experience. OpenAI upgraded deep research in February 2026 to run on GPT-5.2 with several improvements, and that version continues unchanged.

The distinction matters because some users built workflows around the legacy mode’s specific behavior. If your deep research prompts have been working fine recently, you’re already on the new version and won’t notice any change.

What Changed Between Legacy and Current Deep Research

FeatureLegacy (being removed)Current (staying)
ModelSpecialized o3GPT-5.2 based
Site restrictionNot availableCan limit to specific sites
MCP integrationNot availableConnect to external data sources
Progress trackingBasicReal-time with ability to interrupt
Report UIStandardImproved formatting and layout
Follow-upsStart new researchRefine mid-research with prompts

The current version is better in every measurable way. The legacy removal is essentially OpenAI cleaning up old infrastructure — the ChatGPT model picker was recently simplified to just Instant, Thinking, and Pro modes with a Configure menu for power users. Removing legacy deep research is part of that simplification.

Who This Affects

You won’t notice if: You’ve been using deep research normally through the ChatGPT interface. The button, the experience, and the output quality remain the same.

You might be affected if:

  • You built automated pipelines or scripts that specifically called the legacy deep research model
  • You preferred the legacy mode’s specific behavior for certain types of queries
  • You use the macOS app — the Deep Research button has been removed from the interface, and you now access it by typing /deep instead

Your data is safe: All historical deep research conversations and generated reports remain accessible. Nothing is being deleted.

Usage Limits (Unchanged)

Deep research limits remain the same across all plans:

PlanDeep Research Queries/MonthLightweight Queries
Free5 lightweight only5
Plus ($20/mo)25 total15 lightweight
Team25 total15 lightweight
Enterprise25 total15 lightweight
Pro ($200/mo)250 total125 lightweight

If you’re hitting these limits regularly, that’s where alternatives become relevant.

Deep Research Alternatives Worth Knowing

ChatGPT’s deep research isn’t the only option for AI-powered research. Here’s how the main alternatives compare:

Claude (Anthropic)

Claude doesn’t have a branded “deep research” feature, but its extended thinking mode and 200,000-token context window make it strong for research tasks. You can upload entire PDFs, reports, or datasets and ask Claude to analyze them in depth. Where Claude shines: precision on complex documents, careful reasoning, and fewer hallucinations on nuanced topics.

Best for: Analyzing documents you already have. Less suited for web-browsing research since Claude doesn’t browse the internet by default.

Price: Pro plan at $20/month.

Gemini (Google)

Gemini’s deep research mode leverages Google’s search infrastructure — which gives it a natural advantage for web-based research. The 2.5 update improved its ability to synthesize information from multiple sources. It integrates tightly with Google Workspace, so if you’re already in Docs/Sheets/Slides, research flows directly into your workflow.

Best for: Web research that needs to pull from many sources. Especially strong when you need current, real-time information.

Price: Free tier available. Advanced plan at $19.99/month.

Perplexity

Perplexity is built specifically for research — it’s not a general-purpose chatbot with research bolted on. Every response comes with inline citations. Its deep research mode completes most tasks in under 3 minutes and scores well on industry benchmarks.

Best for: Quick, citation-heavy research. Academic work, fact-checking, competitive analysis.

Price: Free tier available. Pro at $20/month (200 Pro searches/week, 20 deep research queries/month).

Important note on Perplexity: They recently reduced Pro search limits from 600 to 200 per week, and deep research queries from 50 to 20 per month. If you’re a heavy user, factor that in.

Quick Comparison

ToolApproachWeb BrowsingSpeedCitationsBest Use Case
ChatGPT Deep ResearchAutonomous agent, 5-30 minYes (extensive)Slow but thoroughEnd of reportComprehensive reports
Claude Extended ThinkingDeep reasoning, no browsingNo (upload docs)MediumIn-contextDocument analysis
Gemini Deep ResearchGoogle Search poweredYes (Google)MediumInlineWeb research + Workspace
Perplexity Deep ResearchResearch-first designYes (focused)Fast (~3 min)Inline, every claimQuick research, citations

What to Do Right Now

If you’re a casual user: Nothing. Open ChatGPT tomorrow, use deep research as normal. The experience hasn’t changed.

If you built workflows on the legacy mode: Test your prompts with the current deep research today. The GPT-5.2 version handles most queries better, but behavior differences exist — especially in how it structures multi-source reports.

If you want to explore alternatives: Try Perplexity for speed and citations, Gemini for web research, or Claude for document analysis. Each has a free tier or trial. No single tool is best at everything — the right choice depends on whether you need web browsing, document analysis, speed, or citation quality.

If you’re frustrated with limits: ChatGPT’s 25 queries/month on Plus is the most restrictive of the bunch. Perplexity gives 20 deep research queries plus 200 regular Pro searches. Gemini’s limits are more generous for general use. If deep research is a core part of your workflow, compare pricing and limits before committing to one platform.


OpenAI’s legacy deep research mode is being removed on March 26, 2026. The current deep research experience remains unchanged. All historical conversations and reports remain accessible.

Sources:

Related courses: AI Fundamentals | Prompt Engineering | ChatGPT vs Claude

Build Real AI Skills

Step-by-step courses with quizzes and certificates for your resume