Deep research reports
Testing nowWhy it's hard
Confident but wrong; weak sources
What we test
5 questions with known answers, every citation checked
best deep research ai · 深度研究哪个AI好
Area 05 · 7 jobs
Best AI for research and study, job by job
We don't rank models in general. We test them one job at a time: the same inputs and settings for every model, graded blind, with the cost of each usable result. 2 of these jobs are being tested now.
Testing now
Academic literature review and citations
Testing nowWhy it's hard
Invented or wrong citations; integrity rules
What we test
3 topics; real-paper rate and citation accuracy
best ai for literature review · AI写论文文献综述
Coming in month 2
Reading and summarizing papers and PDFs
Month 2Why it's hard
Misreads tables and figures
What we test
5 papers, 20 factual questions
ai pdf summarizer
Market and competitor research
Month 2Why it's hard
Stale data, missing players
What we test
3 markets, facts checked
ai market research tool
Math and homework, step by step
Month 2Why it's hard
Wrong steps that look right
What we test
30 problems, K-12 to university
best ai for math homework
Coming in month 3
Exam prep and tutoring
Month 3Why it's hard
Can't adapt to level
What we test
3 subjects
ai tutor
Current-events accuracy
Month 3Why it's hard
Out-of-date answers
What we test
20 recent-fact questions
most accurate ai for news
Models in this area
The field as of 29 Oct 2026. Prices are list prices.
| Model | Latest version | List price | Mainland China |
|---|---|---|---|
| Claude Opus 5.5Anthropic | 22 Sep 2026 | $4 / $20 | No |
| GPT-6 familyOpenAI | Astra 3 Sep; in ChatGPT for all 7 Oct; 6.1 Sol 29 Sep; 6.1 Terra 26 Oct | Astra $10/$50, Sol $2/$10 | No |
| Gemini 4 ArgonGoogle | GA 13 Oct 2026 | $2/$10 offer, then $4/$20 | No |
| DeepSeek V4.1 Flash-1017 | 17 Oct 2026 | ~$0.15 / $0.60 | Yes |
| Kimi K3.1 ProMoonshot | 23 Oct 2026 | $2 / $10 | Yes |
| Doubao 2.1 ProByteDance | 23 Jun 2026 | ¥6 / ¥30 | Yes |
How we choose and grade: our methodology.