Best AI research tools
Checked Sep 28, 2026 · ranked by AI League points from public sources, not by who pays us (nobody does).
Tools that search the web or scientific papers and write cited answers and research reports. Quality comes from blind-vote scores of each app's search model. Ranked by our AI League points: quality, ease, free plan, value for money and language support, each from public sources listed on the app pages.
- 01Best value, built into Google appsFree plan70+ languagesWeb · iPhone · Android
- Cheapest paid plan in this task
- Strong free plan including Deep Research
- Best models have varying access on the free plan
- Google account required
- 02Best all-round AI assistantFree plan60+ languagesWeb · iPhone · Android · Desktop
- Most complete toolset: images, voice, files, deep research
- Generous free plan
- Best features need Plus ($20)
- Ads on Free and Go in some markets
- 03Best for answers with sourcesFree plan52+ languagesWeb · iPhone · Android · Desktop
- Cites sources for every answer
- Pro lets you choose OpenAI, Claude or Gemini models
- Free plan is mostly its own Sonar model
- Paid plan is $20
- 04Best for quick answers from peer-reviewed researchFree planLanguages: not publishedWeb
- Answers grounded in peer-reviewed studies
- Recurring free plan
- Web only
- Deep reviews limited on free
- 05Best for systematic reviews of scientific papersFree planLanguages: not publishedWeb
- Built for academic papers and systematic reviews
- Extracts data from papers into tables
- Pro is expensive for students
- Web only
Standings
| # | Movement | App | Qual | Ease | Free | Value | Lang | Form | Pts |
|---|---|---|---|---|---|---|---|---|---|
| 1 | – | 7.6 | 10.0 | 9.0 | 10.0 | 10.0 | 91 | ||
| 2 | – | 9.3 | 10.0 | 9.0 | 6.2 | 10.0 | 89 | ||
| 3 | – | 5.1 | 10.0 | 9.0 | 2.5 | 10.0 | 69 | ||
| 4 | – | 5.0* | 6.0 | 9.0 | 2.5 | 5.0* | 53 | ||
| 5 | – | 5.0* | 6.0 | 9.0 | 1.0 | 5.0* | 50 |
How points are calculated — Pts = (Quality×30% + Ease×20% + Free plan×15% + Value×20% + Languages×15%) × 10. Quality = blind-vote score of the app's best search model on the LMArena Search leaderboard (1000 = 0, 1275 = 10); not listed = default 5.0. Ease = 2 points each for web, iPhone, Android, starting from a plain-language prompt and starting without paying. Free plan: recurring free allowance 9, one-time trial 5, none 0, minus 1 each for a watermark and low-resolution or non-commercial output. Value = lowest entry price in the task ÷ this app's cheapest monthly plan × 10 (the cheapest paid plan in the task = 10); no public price = default 5.0. Languages by the number documented (50+ = 10).
* Not published: the documented default (5) is used, never a guess. Form = last 5 matchdays: points up · same · down. A new matchday is added only when points change. Each app page lists the sources behind its scores.