I need a researched, cited answer, not a quick chat reply

+ The broadest feature set in one place: Deep Research, code-executing data analysis on uploaded spreadsheets, voice mode and a huge library of custom GPTs.
− Free and Plus tiers hit message caps on the best models at busy times, and answers can be confidently wrong without citations unless you ask it to search.

+ Deep Research produces long cited reports from dozens of sources, and it reads your Gmail, Drive and Calendar directly when you allow it.
− Quality swings between Flash and Pro models, and it refuses or hedges on some topics more than competitors.

+ Every answer comes with clickable sources, and Pro lets you switch between frontier models from OpenAI, Anthropic and Google.
− Citations sometimes do not actually support the sentence they are attached to, so you still have to click through for anything important.

+ The Consensus Meter gives an at-a-glance split of studies supporting or opposing a claim, with each paper linked.
− Only indexes academic literature, so it is useless for questions the scientific record does not cover, and paper summaries can oversimplify findings.

+ Pulls real-time posts from X into answers, which makes it the fastest way to research breaking news and public sentiment.
− Fewer guardrails and a reliance on X content mean more unverified claims and off-color outputs than other assistants.

+ The Copilot sidebar explains equations, tables and jargon in a paper you are reading without leaving the PDF.
− The free tier is limited and the interface pushes upgrades aggressively.

+ Turns dozens of papers into a structured table of extracted data columns, which is exactly what a literature review needs.
− Free credits run out quickly, and extracted fields need spot-checking against the source PDF.