All tasks / Comparisons
ChatGPT vs Perplexity
Compared only on public benchmark results, across the 4 tasks where both are listed: ChatGPT is a tier higher in 1 of 4 shared tasks, they share a tier in 2.
ChatGPT
OpenAI's chat assistant on web, desktop and mobile, built on its GPT model family.
- Vendor
- OpenAI
- Kind
- Apps
- Runs
- Cloud
- Price
- Free tier · from $20/mo
- Skill level
- Beginner: works out of the box
- Open source
- No
- Last verified
- 18 Oct 2026
Perplexity
Search-based answer engine on web, desktop and mobile that cites sources; paid plans can pick third-party models.
- Vendor
- Perplexity AI
- Kind
- Apps
- Runs
- Cloud
- Price
- Free tier · from $20/mo
- Skill level
- Beginner: works out of the box
- Open source
- No
- Last verified
- 18 Oct 2026
Task by task
| Task | ChatGPT | Perplexity | Higher tier |
|---|---|---|---|
| Chat & assistants Apps | Svia GPT-6.1 Sol · Cloud | Svia Gemini 4 Argon · Cloud | Same tier |
| Search & research Apps | A48 on DeepResearch Bench (RACE, GPT-5.5 judge) · Cloud | B43 on DeepResearch Bench (RACE, GPT-5.5 judge) · Cloud | ChatGPT |
| Image understanding Apps | Svia GPT-6.1 Sol · Cloud | Svia GPT-6.1 Sol · Cloud | Same tier |
| Browser & computer agents Agent apps & AI browsers | –No public result yet · Cloud | –No public result yet · Cloud | — |
A tier spans a range of scores, and small gaps on a leaderboard are often within its margin of error, so only a different tier counts as ahead. How we rank.
Same benchmark, both results
| Benchmark | ChatGPT | Perplexity | Snapshot |
|---|---|---|---|
| DeepResearch Bench (RACE, GPT-5.5 judge) RACE overall score (0-100; report quality judged by GPT-5.5 relative to reference reports on 100 research tasks), higher is better | 48 | 43 | 11 Oct 2026 |
| DeepResearch Bench II (GPT-5.5 text-only eval) TotalScore: expert-rubric pass rate (%) over 130 scored research tasks (tasks 112 and 114 temporarily excluded), judged text-only by GPT-5.5, higher is better | 43.1% | 33.6% | 19 Oct 2026 |
| Mind2Web 2 (agentic search, private test set) Success rate: share of 130 long-horizon, time-varying agentic search tasks whose answer gets a perfect root score from the Agent-as-a-Judge rubric tree (leaderboard's default sort), higher is better | 28% | 15% | 19 Oct 2026 |