Introduction
A research brief can look finished long before it is trustworthy. The sentences flow. The citations sit neatly at the bottom. Then someone opens a source and discovers it doesn’t quite say what the brief claims. We evaluated Claude with that gap in mind: the distance between an answer that reads well and one a reader can actually check.
Claude is Anthropic’s family of conversational AI models. It handles writing, analysis, coding, and Research. In the company’s 2021 founding announcement, Dario Amodei is named chief executive and Daniela Amodei president. Claude first worked through a closed alpha with Notion, Quora, and DuckDuckGo, then became more broadly available in Anthropic’s March 14, 2023 Claude announcement. Its roots are in a general, steerable model designed for reliable work.
How many people pay for it? Anthropic has not disclosed an exact current subscriber count. In a March 28, 2026 subscription report, an Anthropic spokesperson said paid subscriptions had more than doubled during 2026, without providing either the initial number or the new total. That report’s consumer-user estimates are estimates. Anthropic’s separate figure of more than 300,000 business customers counts businesses, not paid consumer subscriptions. Neither number fills the gap in the subscriber total.
OTeam’sm’s hands-on Research and experiments conducted through September 25, 2026 put Claude third of six, at 8.73/10. We examined how an assistant plans a complex question, finds and compares sources, holds on to constraints, works across long documents, exposes evidence, and handles revisions. Claude was especially strong at reasoning. Whether that strength makes it your best choice depends on the work waiting on your desk. Our ranking is a Find Premium AI study, not an official market order; different tasks, prompts, and weights could reasonably place Claude first.
Performance
Claude’s result is easiest to understand as two numbers side by side: 9.4 for AI Intelligence and 8.7 for Information and Research Quality. It can think through a complicated assignment exceptionally well. The evidence gathering and checking still need a human partner. We weighted quality more heavily than any other factor because a persuasive explanation is only as useful as the material beneath it.
The numbers behind the judgment
The scorecard uses a weighted 10-point scale: AI Intelligence 15%; Speed 10%; Information and Research Quality 25%; Prompt Accuracy 20%; User Rating 10%; Review Confidence 5%; Features and Usability 15%.
| Evaluation Factor | What It Measures | Weight |
|---|---|---|
| AI Intelligence | Reasoning, planning, tools, context, adaptation, and analysis. | 15% |
| Speed | Consistent delivery time for verified research outputs, including iterations and revisions. | 10% |
| Information and Research Quality | Source discovery, selection, freshness, citations, synthesis, gaps. | 25% |
| Prompt Accuracy | Scope adherence, completeness, accuracy, uncertainty, and avoiding unsupported claims. | 20% |
| User Rating | Weighted public satisfaction from credible review platforms | 10% |
| Review Confidence | Editorial calibration of verification, relevance, recency, diversity, and product fit. | 5% |
| Features & Usability | Search, sources, files, citations, workflows, accessibility, and limits. | 15% |
The column letters mean Intelligence, Speed, Quality, Prompt accuracy, User rating, Review confidence, and Features and usability. Scite AI’s overall 7.84 is reproduced as printed in the supplied scorecard. Its displayed component scores and weights instead add up to approximately 7.79. That discrepancy doesn’t change Claude’s position, but readers deserve to see it.
Think of a brief with a long PDF, several competing sources, a strict date window, and a request to show where claims came from. The 9.4 Intelligence result speaks to planning the subquestions, carrying context through the document set, choosing tools, and revising when evidence disagrees. The 9.1 Prompt Accuracy result reflects how well scope, dates, exclusions, source rules, format, and uncertainty survive the trip from prompt to final answer. For a literature brief or policy memo, those details are the assignment.
The 8.7 Quality score brings the necessary restraint. Source coverage, freshness, primary documents, and claim-level citation fidelity are separate jobs. Claude can help discover and synthesize information, but you still need to open consequential links, read the passage behind a citation, and look for the source it missed. Anthropic’s research guidance says the Research feature performs successive searches, follows open questions, and can draw on connected internal context. The same guidance says Research uses the ordinary conversation limits and may use them faster as it retrieves more material. Useful reach comes with real friction.
That friction is part of the 7.7 Speed result. We measured the time and consistency of reaching a usable, checkable report, including searches, retries, and revisions. We did not time a single response with a stopwatch. A careful multi-step answer can justify the wait. For repeated quick lookups, though, Perplexity’s 9.2 speed score offers a noticeable advantage, and ChatGPT’s broader tool workflow may feel more immediate. Claude’s method pays off most when there is something substantial to untangle.
The model and the working limits
The newest model release by our cutoff was Claude Opus 5.5, announced September 22. Anthropic says Opus is available on its platform and major cloud platforms. The company’s Fable release calls Fable 5.1 generally available and reserves Mythos 5.1 for trusted-access programs in sensitive cybersecurity and life-sciences work. For ordinary information and Research, Fable 5.1 is our practical reference for long-running knowledge work; Opus 5.5 is newer and makes sense when the task warrants its higher usage cost.
Anthropic says Opus 5.5 is roughly on par with Fable 5.1 for most work, costs 40% less to run than Opus 5, and generates output more than 30% faster than Opus 5. For Fable 5.1, Anthropic reports 60.9% on Humanity’s Last Exam without tools, 65.0% with tools, and 52.6% on Terminal-Bench-Science 0.1 in its model announcement. These are Anthropic’s controlled benchmark and cost claims. They do not certify the accuracy of a particular research brief or replace our product-level scorecard.
The file workflow has boundaries, too. Anthropic’s file guidance covers PDFs, DOCX, CSV, and other common formats, with limits on size and files per chat. Very long PDFs receive different visual versus text-only processing. Its context guidance describes up to one million tokens on supported paid models. Capacity is helpful; it does not promise equal attention to every paragraph. Projects, retrieval, and an explicit evidence ledger can make a long assignment easier to manage.
Four checks belong beside any buying decision. Accuracy: Anthropic’s accuracy guidance acknowledges factual errors, misleading citations, and missing information. Freshness: web search and Research help, but important claims still need direct verification. Access: web, desktop, mobile, and Claude Code share usage, with rolling five-hour windows and extra paid-plan caps; extensive Research sessions can reach a limit quickly. Privacy: consumer settings and commercial arrangements differ under Anthropic’s privacy policy. Check training and feedback choices before putting confidential Research in a consumer account.
What the public stars actually tell us
Public reviews answer a useful question: How do people feel about using the product? They do not answer whether a research claim is correct.
For Claude, we prioritize Google Play as a mass-market user-experience signal. A Claude-specific snapshot dated September 22, 2026 records 4.5/5 from approximately 734,000 reviews of the official Claude by Anthropic Android app. We chose it for the visible five-star scale, clear product identity, and unusually large product-specific sample. For comparison, the same snapshot lists the Apple App Store at about 266,000 ratings, G2 at 468 reviews, TrustRadius at 349 reviews and ratings, Gartner Peer Insights at 97 ratings, and Capterra at 58 reviews. Professional sites add a different, smaller view of business use. Trustpilot’s 1.5/5 from 1,963 reviews offers a sharply different view from another self-selected group. The highest rating did not determine our source choice.
The supplied six-product ratings panel did not specify when it was captured. Its comparator figures appear below. Claude’s row uses the dated Claude-specific snapshot; the panel itself showed 4.4/5 from 753K+ reviews for Claude. Timing or a different Google Play listing surface may explain the difference; we cannot establish its historical capture date from that panel. We checked the linked listings for product identity and scales on September 26, 2026. The figures should not be mistaken for a single synchronized market survey.
| Product | Source used | Public Rating | Votes & Reviews |
|---|---|---|---|
| ChatGPT | Google Play Store | 4.5/5 | 61.8M reviews |
| Gemini Deep Research | Google Play Store | 4.4/5 | 46.9M reviews |
| Claude | Google Play Store | 4.4/5 | 753K+ reviews |
| Perplexity | G2 | 4.4/5 | 368+ reviews |
| Gemini Notebook | Google Play Store | 4.6/5 | 325K+ reviews |
| Scite AI | Trustpilot | 4.4/5 | 274+ reviews |
We prefer listings where readers can inspect the scale, count, product, and review process. Google Play documents automated and human review moderation, plus regional and device variation. G2 describes identity and conflict checks and human review screening. Trustpilot’s rating method accounts for review recency and volume and describes detection and human review. None offers a universal sample: Google Play favors Android users, G2 draws business buyers, and Trustpilot attracts people motivated to post. Other platforms can add useful perspectives. These star ratings provide context for user experience; they are not the six products’ official scores, ranks, or research-accuracy tests.
Comparisons Before You Buy
Start with your research job, then look at the ranking. Claude’s 8.73/10 puts it third, 0.01 behind Perplexity and 0.14 behind Gemini Deep Research. ChatGPT leads at 8.94, combining strong quality and prompt accuracy with the highest Features and Usability score in the group. Gemini Deep Research beats Claude on quality and features. Perplexity wins on speed for current, cited lookups. Gemini Notebook is a natural fit for studying an already chosen set of notes and documents. Scite AI is the specialist for citation context and literature auditing, but it has lower intelligence, quality, and features as a general research assistant.
| Product | Overall Score | AI Intelligence | Speed | Information and Research Quality | Prompt Accuracy | User Rating | Review Confidence | Features & Usability | Best For |
|---|---|---|---|---|---|---|---|---|---|
| ChatGPT | 8.94 | 9.3 | 7.1 | 9.2 | 9.1 | 9.4 | 7.5 | 9.3 | Complex sourced reports from files |
| Gemini Deep Research | 8.87 | 9.1 | 7.5 | 9.3 | 8.8 | 9.1 | 7.5 | 9.2 | Google Workspace integrated research briefs |
| Claude | 8.73 | 9.4 | 7.7 | 8.7 | 9.1 | 9.3 | 7.5 | 8.3 | Nuanced synthesis of long documents |
| Perplexity | 8.72 | 8.5 | 9.2 | 8.8 | 8.6 | 9.2 | 7.5 | 8.7 | Fast current cited information lookups |
| Gemini Notebook | 8.45 | 8.1 | 8.5 | 8.6 | 8.5 | 9.5 | 6.2 | 8.5 | Study selected documents and notes |
| Scite AI | 7.84 | 7.3 | 8.2 | 7.8 | 7.8 | 9.1 | 6.0 | 7.7 | Citation context and literature auditing |
Those differences matter more than a podium. Working through a messy document collection and writing a nuanced brief? Claude’s reasoning and prompt discipline make a compelling case. Moving among files, tools, and output formats? ChatGPT is the safer generalist. Already living in Google Workspace? Gemini Deep Research has the integrated brief workflow. Need a trail of current web sources quickly? Perplexity is well suited to it. The lower-ranked choice may be the better choice for your work.
The subscription decision
For the regular individual researcher, choose Claude Pro. Anthropic’s pricing page, checked September 26, 2026, lists US$20 per month or US$200 billed annually. Annual billing works out to US$16.67 per month and saves US$40 a year compared with twelve monthly payments. Prices exclude applicable tax; region and app store can change what you pay. Anthropic does not publish a dated price archive, so these are the closest public figures to the September 25 cutoff; check the price again before buying.
Pro adds more usage, Projects, additional model access, web search, and paid-plan Research. That mix suits several substantial research jobs a month without paying for all-day capacity. Research uses the same pool as ordinary chat and can hit a cap during a long, multi-source job. Paid users can access Opus 5.5; Fable access is described as usage credits, rather than unlimited included capacity.
Your workload may point elsewhere:
- Free includes web search and suits occasional questions, quick summaries, and trying Claude. Lower usage makes it a poor fit for repeated, document-heavy work.
- Max 5x costs US$100 per month; Max 20x costs US$200 per month. They target people who use Claude throughout the day and value fewer interruptions—both bill monthly, with no annual Max discount. The five- and twenty-times figures describe per-session usage relative to Pro, alongside higher output limits and priority access. Rolling or weekly caps still apply. More capacity does not guarantee better answers.
- Team requires at least two seats and adds centralized billing, organization features, and more usage than Pro. Standard is US$25 per seat per month on monthly billing or US$20 per seat per month on annual billing: US$300 versus US$240 per seat per year. Premium is US$125 monthly or US$100 annually per seat: US$1,500 versus US$1,200 per seat per year. Its allowance is five times Standard. Give that capacity to the people whose research volume calls for it; the whole team need not be on Premium.
- Enterprise uses an annual seat-plus-usage arrangement. The listed base is US$20 per seat per month plus usage at API rates, with advanced administration, audit logs, SCIM, role-based access, and custom retention controls. It earns consideration when procurement, identity, retention, and governance of internal sources are part of the assignment. A small group should compare likely models, and Research should use Teams before committing.
Annual Pro and Team plans reward steady use over most of a year when the upfront bill is comfortable. Monthly Pro suits seasonal work, an uncertain project, or a trial of Claude’s long-document workflow. Max buys capacity, not an accuracy upgrade. Whatever the plan, someone still needs to check the quotations, dates, calculations, and primary sources before the brief goes any further.
Conclusion
Claude’s strongest move is to give complicated material a shape without losing the instructions that shaped it. Its 9.4 Intelligence and 9.1 Prompt Accuracy results speak to planning, context, and careful execution. The 8.7 Quality and 7.7 Speed scores mark the work left for the researcher: verify sources and allow time for the multi-search process when the question demands it.
Our result ranks third at 8.73/10, behind ChatGPT and Gemini Deep Research, based on the tasks and weights we used through September 25, 2026. It is not an official market ranking. For recurring individual Research, Pro is the recommendation. Free fits light use; Max supports sustained volume; Team supports shared work; Enterprise serves governance needs. Choose Claude when a careful, coherent brief is worth more to you than the fastest first draft.
FAQ
What does Claude’s 8.73/10 overall score really say about its research performance?
Claude’s score represents end-to-end research performance, not reasoning ability alone. It led the comparison in AI Intelligence at 9.4/10 and scored 9.1/10 for Prompt Accuracy, showing strong planning, context handling, instruction-following, and revision.
Its overall score dropped because of lower results in Information and Research Quality (8.7), Speed (7.7), and Features and Usability (8.3). Because research quality carried the largest weighting at 25%, these differences mattered. Claude finished 0.14 points behind Gemini Deep Research and 0.01 points ahead of Perplexity.
For a complex, constrained brief, Claude’s reasoning strengths may matter more than its third-place ranking. For rapid current-information searches, speed and source workflow may matter more.
What should a researcher verify after Claude produces a polished report?
The most important checkpoint is the gap between good reasoning and verified evidence. Claude can organize a difficult question, compare sources, and produce a convincing synthesis while still missing a relevant source or using a citation that only partially supports a claim.
The article’s 8.7/10 research-quality score evaluates source coverage, freshness, primary-source quality, claim-level citation faithfulness, synthesis, and explicit acknowledgment of gaps. This is a strong result, but it does not guarantee citation accuracy.
Before relying on the report, researchers should open important links, read the cited passage, confirm the publication date and source authority, check figures and quotations, compare conflicting evidence, and distinguish documented facts from Claude’s inferences. Claude can accelerate research; a human still needs to approve the evidence.
What type of research assignment is Claude particularly well suited to?
Claude is strongest when the task involves complex reasoning, long documents, competing sources, strict instructions, and several rounds of revision. It can help break a difficult question into subquestions, preserve constraints such as date ranges and exclusions, carry context across a document set, and revise an answer when evidence conflicts.
This makes it a strong choice for a policy memo, literature brief, or research report that needs careful organization and nuanced explanation. Perplexity may be better for fast, current, cited lookups; ChatGPT for a broader tool workflow; Gemini Deep Research for a Google-integrated report; and Scite AI for citation-focused literature auditing.
Is Claude’s Research mode sufficient for a current, source-heavy research project?
Research mode can perform successive searches, follow unresolved questions, and use connected internal context. That makes it useful for multi-source assignments requiring more than a single answer. Supported paid models can also handle context windows of up to one million tokens, depending on the plan and model.
There are practical limits. Research uses the same conversation allowance as regular Claude use, so a long session can quickly consume the available limit. A large context window also does not guarantee equal attention to every paragraph. For a dependable result, users should provide a clear evidence structure, maintain an evidence ledger, and verify consequential claims against the original sources.
Which Claude plan fits different research workloads, and what does paying more actually provide?
Claude Pro is the article’s recommendation for most individual researchers. It costs US$20 per month or US$200 annually an effective US$16.67 per month when billed annually, saving US$40 over twelve monthly payments. Pro adds higher usage, Projects, web search, additional model access, and paid-plan Research.
Free suits occasional questions and light summaries. Max 5x costs US$100 per month, while Max 20x costs US$200 per month; these plans are for people who use Claude heavily and need more capacity, priority access, and higher limits. They do not guarantee more accurate research.
Team is designed for shared work and requires at least two seats. Enterprise is appropriate when administration, identity management, audit logs, retention, and governance of internal sources are central requirements. The right upgrade depends on workload, collaboration, and governance needs, not on an assumption that a more expensive plan automatically produces better evidence.
