All tasks / Comparisons
ChatGPT vs Gemini
Compared only on public benchmark results, across the 11 tasks where both are listed: ChatGPT is a tier higher in 2 of 11 shared tasks, Gemini in 3, they share a tier in 3.
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
Gemini
Google's chat assistant on web and mobile; plan tier sets access to Flash-Lite, Flash, Pro and Gemini 4 Argon models.
- Vendor
- Kind
- Apps
- Runs
- Cloud
- Price
- Free tier · from $4.99/mo
- Skill level
- Beginner: works out of the box
- Open source
- No
- Last verified
- 18 Oct 2026
Task by task
| Task | ChatGPT | Gemini | 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 | A50 on DeepResearch Bench (RACE, GPT-5.5 judge) · Cloud | Same tier |
| Writing Apps | Avia GPT-6.1 Sol · Cloud | Svia Gemini 4 Argon · Cloud | Gemini |
| Translation Apps | –No public result yet · Cloud | –No public result yet · Cloud | — |
| Documents & OCR Apps | –No public result yet · Cloud | –No public result yet · Cloud | — |
| Data analysis Apps | C45.5% on SpreadsheetBench (V1 full, 912 tasks) · Cloud | Bvia Gemini 3.1 Pro (Preview) · Cloud | Gemini |
| Image generation Apps | Svia GPT Image 2.5 · Cloud | Avia Nano Banana 2.1 · Cloud | ChatGPT |
| Image editing Apps | Svia GPT Image 2.5 · Cloud | Avia Nano Banana 2.1 · Cloud | ChatGPT |
| Image understanding Apps | Svia GPT-6.1 Sol · Cloud | Svia Gemini 3.8 Flash · Cloud | Same tier |
| Voice agents Apps | Avia GPT-Live 1 · Cloud | Svia Gemini 3.8 Live · Cloud | Gemini |
| 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 | Gemini | 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 | 50 | 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% | 41.5% | 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% | 18% | 19 Oct 2026 |