Copilot vs Gemini vs Perplexity: 1B Users, $325 Gap [2026]

Microsoft Copilot just crossed a milestone that used to belong to search engines and social networks: Google confirmed on August 11, 2026 that the Gemini app now serves more than 1 billion monthly active users, up from 950 million just three weeks earlier at Alphabet’s Q2 2026 earnings call. Meanwhile Perplexity AI, the citation-first “answer engine,” reported roughly 100 million monthly active users in April 2026, more than double its 45 million at the end of 2025. And Microsoft Copilot, freshly running on OpenAI’s GPT-5.6 family since July 9, 2026, remains locked inside a fight over what “active user” even means, with figures ranging from 33 million to 420 million depending on which surface Microsoft is counting.

These three products no longer compete on novelty. They compete on where you already work. Copilot lives inside Word, Excel, and Teams. Gemini lives inside Docs, Sheets, and Gmail. Perplexity lives wherever you’d otherwise open a search tab. This comparison breaks down the current models, pricing, benchmarks, and real deployment costs behind all three as of August 22, 2026, so you can figure out which one actually earns a seat in your workflow — and which one is just riding its parent company’s distribution.

·

Google · Preferred Sources

Don't miss new tech stories on Google

Add Tech Insider once in the Google app and our stories appear in your news suggestions.

Add Now

Why Compare Copilot, Gemini, and Perplexity in 2026

Most “best AI chatbot” roundups pit ChatGPT against Claude against Gemini, because those three ship the flagship reasoning models everyone benchmarks. Copilot, Gemini, and Perplexity is a different comparison entirely: it’s a fight over distribution and integration, not raw model IQ. Copilot doesn’t train its own frontier model — it wraps OpenAI’s GPT-5.6 family inside the Microsoft 365 suite. Perplexity doesn’t train a general-purpose chat model either — it built Sonar, a search-tuned model, and lets Pro subscribers pick from GPT-5.x, Claude, Gemini, or Grok underneath. Gemini is the outlier: Google trains its own frontier model (Gemini 3 Pro) and ships it as both a standalone app and a Workspace layer.

That structural difference matters more than any leaderboard score. If you’re choosing a chatbot for a Microsoft 365 shop, Copilot’s GPT-5.6 backend is close to irrelevant — what matters is whether it can competently rewrite an Excel formula or summarize a Teams thread. If you’re a researcher who needs sourced, verifiable answers, Perplexity’s citation-first design beats a marginally smarter model that doesn’t show its work. And if your organization already lives in Google Workspace, Gemini’s native Docs and Sheets integration is worth more than a few extra points on GPQA Diamond. This piece treats all three fairly on their own terms: integration depth, pricing reality, and benchmark performance where it’s actually documented.

Microsoft Copilot in 2026: GPT-5.6, Microsoft 365, and a Confusing User Count

Microsoft 365 Copilot switched to GPT-5.6 as its preferred model on July 9, 2026, according to both Microsoft’s own Copilot documentation and OpenAI’s simultaneous announcement. The rollout covers Word, Excel, PowerPoint, Copilot Chat, and the newer “Cowork” agent surface, replacing earlier GPT-5.2 and GPT-5.4 “Thinking” deployments that had been staggered through the first half of the year. GPT-5.6 ships in three named capability tiers — Sol, Terra, and Luna — with Sol positioned as the flagship for complex reasoning, coding, and research tasks.

Pricing tells the real story of where Copilot is trying to win. For individuals outside a Microsoft 365 subscription, Copilot Pro is priced separately at $20 per user per month as of May 2026, according to Subkept. Microsoft 365 Copilot Business is priced at $18 per user per month as a promotional rate running through December 31, 2026, reverting to $21 afterward. Copilot Enterprise runs $30 per user per month, billed annually — a price point Computerworld confirmed held steady through July 2026 with no increase — but that’s an add-on cost stacked on top of an existing Microsoft 365 E3 ($36/user/month) or E5 ($57/user/month) license, pushing the effective total to $66–$87 per user per month for large organizations, a range corroborated by a separate Professional Business Directory estimate from March 2026. For a 10,000-person enterprise deploying Copilot at full scale, that works out to roughly $3.6 million a year in incremental spend, according to enterprise-decision analysis from SyncRivo published May 4, 2026. Microsoft has also begun layering metered billing on top of those flat seat prices: in June 2026 it introduced Copilot Credits, charging $0.01 per credit for agent runs and premium features that exceed a plan’s included allotment, giving IT budgets a usage-based lever alongside the per-seat subscription.

Copilot’s context window is murkier than its competitors’. Microsoft doesn’t publish an official conversation-length limit for Copilot Chat in Microsoft 365; independent trackers estimate around 128K tokens based on the GPT-4.1-class models used as defaults in Copilot Studio, while users report practical, unofficial conversation limits closer to 25,000–40,000 tokens before responses degrade. That gap between documented capability and lived experience is a recurring complaint in enterprise IT forums.

User counts are where Copilot gets genuinely confusing. Microsoft’s own disclosures put “Copilot family” monthly active users at roughly 150 million, but third-party trackers report anywhere from 33 million (narrow app/web/Windows definitions) to 420 million (all surfaces combined, including Bing and Windows Copilot), according to a July 2026 MAU analysis from RocketBlue. Separately, Microsoft says Copilot processes an estimated 12 billion queries per month as of Q1 2026. Whatever the true active-user figure, Copilot’s paid-seat number is more concrete: over 30 million paid Microsoft 365 Copilot seats as of Q4 FY2026.

Google Gemini in 2026: 1 Billion Users and the Gemini 3 Family

Gemini’s growth curve in 2026 is the standout story of this comparison. Google confirmed on August 11, 2026 that the Gemini app surpassed 1 billion monthly active users, a jump from 950 million just weeks earlier when CEO Sundar Pichai disclosed the figure on Alphabet’s Q2 2026 earnings call, as reported by TechCrunch on July 23, 2026. That’s roughly 6.6x Copilot’s most Microsoft-favorable “family” figure and 10x Perplexity’s user base, though it’s worth noting Gemini’s count benefits enormously from default placement on every Android phone and inside the Google app.

The flagship model behind that growth is Gemini 3 Pro, which Google launched November 18, 2025 with benchmark scores that led the field at the time: 91.9% on GPQA Diamond and 91.8% on MMLU, according to Google’s official model announcement and corroborated by VentureBeat’s coverage the same day. That lead has since widened: an updated Gemini 3.1 Pro pushed the GPQA Diamond score to 94.3% by June 2026, compared with 92.4% for GPT-5.2, according to Fast.io’s benchmark tracking. The API side of the product line has continued iterating rapidly: Google Cloud launched Gemini 3.5 Flash on May 19, 2026 at $1.50 per 1M input tokens and $9 per 1M output tokens, before Google DeepMind leapfrogged it in September 2026 with Gemini 3.8 Flash, undercutting that price sharply to $0.75 per 1M input tokens and $3.75 per 1M output tokens through at least December 2026, while carrying the same roughly 1,048,576-token (about 1M) input context window and around 64,000–65,000 output tokens. Google raised the flagship bar again that same month with Gemini 4 Argon, priced at $2 per 1M input tokens and $10 per 1M output tokens, which Google reported scoring 77.9% on the DeepSWE v1.1 coding benchmark — the clearest signal yet that Gemini’s model cadence is now outpacing both Copilot’s and Perplexity’s underlying suppliers.

Consumer-facing context windows in the Gemini app scale with subscription tier, and Google documents this explicitly on its Gemini Apps limits page: users without a paid AI plan get a 32K-token context window, Google AI Plus subscribers get 128K tokens, and Google AI Pro and AI Ultra subscribers get the full 1M-token window. Pricing runs $4.99/month for AI Plus (bundled with roughly 400GB of Drive storage in several regional promotions) and $19.99/month for AI Pro (5TB storage, full 1M context, Gemini 3 Pro access) — the tier formerly marketed as Gemini Advanced before Google folded it into the Google AI Pro brand in July 2026, according to Techpresso — plus roughly $250/month for AI Ultra, which adds “Deep Think” reasoning mode with an effective usable context closer to 192K tokens in practice, according to independent usage analysis.

Gemini’s business differentiator is native Workspace integration — Docs, Sheets, Gmail, Meet, and Drive all get Gemini-powered features without leaving the app. Gemini for Google Workspace is bundled starting at the Business Standard tier ($14/user/month base Workspace cost) and above, with a standalone Gemini Enterprise add-on typically negotiated between $18 and $25 per user per month on top of a Workspace Enterprise license, according to SyncRivo’s May 2026 enterprise decision guide — pushing the all-in effective cost to roughly $48–$60 per user per month, per a separate Professional Business Directory estimate from March 2026.

Perplexity AI in 2026: The Answer Engine Built for Citations

Perplexity takes a fundamentally different architectural bet than Copilot or Gemini: instead of shipping one house model for everything, it built a search-and-citation layer called Sonar and lets paying users pick from a rotating menu of frontier models underneath — OpenAI’s GPT-5.x line, Anthropic’s Claude Sonnet and Opus 4.6, Google’s Gemini 3 Pro, and xAI’s Grok, depending on the exact month you check the model picker. Perplexity itself warns that “the exact model names move almost monthly,” which is a candid admission that its core product is the search-and-citation UX, not any single underlying LLM.

Perplexity’s own Sonar models have published, if modest, benchmark numbers. On the SimpleQA factual-accuracy benchmark, Sonar scores an F-score of 0.773 and Sonar Pro scores 0.858, according to a comparison last updated August 16, 2026. A separate API-level benchmark from Parallel.ai (August 19, 2026) put Perplexity’s search API at 92% SimpleQA accuracy at $0.052 per request. On broader reasoning benchmarks, Sonar’s base model scores 47.1% on GPQA Diamond and 68.9% on MMLU Pro — solidly behind Gemini 3 Pro’s 91.9%/91.8%, which underscores that Sonar isn’t built to compete on raw reasoning; it’s built to competently route queries and cite sources fast.

Growth has been Perplexity’s strongest 2026 story. The company reported roughly 100 million monthly active users as of April 2026, more than double the 45 million it had at the end of 2025, per an AI statistics digest published August 10, 2026. Its Comet browser — which went free worldwide in stages through late 2025 and early 2026, reaching iOS on March 18, 2026 — has reported user figures ranging wildly from roughly 3 million to as high as 18 million monthly active users in Q1 2026 depending on the analytics source, a discrepancy neither Perplexity nor independent trackers have fully reconciled.

Pricing is the most straightforward of the three: Perplexity Free includes roughly 5 Pro Searches per day, a heavily capped Deep Research allowance, and no manual model selection (queries route through a default Sonar mode). Perplexity Pro costs $20/month or $200/year and unlocks near-unlimited Pro Search, the full model picker, expanded Deep Research, file uploads, and image generation. On the business side, Perplexity Enterprise Pro runs $40/user/month (or $33.33/month billed annually) with no prerequisite productivity license required — a meaningful contrast to Copilot and Gemini, both of which assume you’re already paying for Microsoft 365 or Google Workspace. Perplexity Enterprise Max, a premium tier aimed at heavy research and video-generation workloads, runs $325/user/month.

Full Specs Comparison: Copilot vs Gemini vs Perplexity

The table below lines up the documented, sourced specs for all three products as of August 22, 2026. Where a figure isn’t officially published, it’s marked as such rather than estimated.

Spec	Microsoft Copilot	Google Gemini	Perplexity AI
Flagship model	GPT-5.6 (Sol/Terra/Luna)	Gemini 3 Pro / Gemini 3.7 Flash	Sonar / Sonar Pro + model picker
Latest model rollout	July 9, 2026	Nov 18, 2025 (Gemini 3 launch)	Rolling updates through 2026
Top context window	~128K tokens (unofficial)	1M tokens (AI Pro/Ultra)	Not officially published
Free-tier context window	Not officially published (~25K–40K observed)	32K tokens	Not officially published (Sonar default)
GPQA Diamond score	Not published for Copilot bundle	91.9%	47.1% (Sonar base)
MMLU / MMLU Pro score	Not published for Copilot bundle	91.8% (MMLU)	68.9% (MMLU Pro, Sonar)
Monthly active users	~150M (Microsoft-disclosed “family”); range 33M–420M by tracker	1 billion+ (Aug 2026)	~100M (April 2026)
Primary integration	Word, Excel, PowerPoint, Outlook, Teams	Docs, Sheets, Gmail, Meet, Drive	Web, Comet browser, API, Slack
Citation-backed answers	Limited inline citations via Bing index	Grounding citations in Deep Research mode	Core feature on every answer
Image/video generation	Designer / Image Creator built in	Imagen and Veo via Gemini app	Available on Pro plan and above
Standalone browser	Edge (Copilot sidebar)	Chrome (Gemini integration)	Comet (dedicated AI browser)
Requires base productivity license	Yes — Microsoft 365	Only for Workspace tiers; free app is standalone	No — standalone at every tier

Pricing Breakdown: Free, Plus, Pro, and Enterprise Tiers

Pricing structures across these three products don’t map cleanly onto each other, which is itself a useful signal about their different business models. Copilot’s consumer access is effectively bundled into a Microsoft account (Bing Chat, Windows Copilot) at no separate charge, with real monetization happening at the Business and Enterprise tiers. Gemini runs a classic freemium ladder from $0 to $250/month. Perplexity keeps it simplest: free, $20/month Pro, or enterprise tiers starting at $40/user/month.

Tier	Microsoft Copilot	Google Gemini	Perplexity AI
Free	Bundled with Microsoft account (Bing/Windows Copilot), capped daily prompts	$0/mo — 32K context, standard features	$0/mo — ~5 Pro Searches/day, Sonar default
Mid tier	—	AI Plus $4.99/mo — 128K context, ~400GB storage	—
Individual Pro	—	AI Pro $19.99/mo — 1M context, 5TB storage	Perplexity Pro $20/mo ($200/yr) — full model picker, near-unlimited search
Premium	—	AI Ultra ~$250/mo — Deep Think, 1M context	—
Business	M365 Copilot Business $18/user/mo (promo to Dec 31, 2026; then $21)	Gemini for Workspace, bundled from Business Standard $14/user/mo	—
Enterprise	$30/user/mo add-on (+ M365 E3 $36 or E5 $57)	~$18–$25/user/mo add-on (+ Workspace Enterprise)	Enterprise Pro $40/user/mo; Enterprise Max $325/mo
Effective enterprise total	$66–$87/user/mo (stacked)	Varies by existing Workspace tier	$40–$325/user/mo (no base license required)

The practical takeaway: Copilot and Gemini both punish you for not already being a customer of their parent suite. If your organization has zero Microsoft 365 or Google Workspace seats today, Perplexity Enterprise’s $40/user/month with no prerequisite license can actually be cheaper than either alternative — a genuinely counterintuitive result for what looks like the “smaller” product on paper.

Benchmark Performance: Reasoning, Search Accuracy, and Adoption Data

Benchmark comparisons across these three products are inherently lopsided, because only Gemini publishes a full reasoning-benchmark suite for its own house model. Copilot inherits whatever OpenAI publishes for GPT-5.6 rather than a Copilot-specific number, and Perplexity’s Sonar is explicitly optimized for search-and-cite tasks rather than academic reasoning benchmarks — which is exactly why its GPQA and MMLU scores trail Gemini’s by a wide margin. One head-to-head coding benchmark does exist across two of the three: on SWE-bench, Fast.io’s June 2026 testing put Gemini at 63.8% versus Copilot at 33.2%, a gap wide enough to suggest Copilot’s edge with developers is more about IDE and Teams integration than raw coding capability. That gap has only grown since: Artificial Analysis found in October 2026 that Google’s Gemini 4 Argon matched OpenAI’s GPT-6 Astra on the Intelligence Index while costing 60% less to run, meaning Gemini now undercuts the field on both capability and price rather than trading one for the other.

Metric	Copilot (GPT-5.6 class)	Gemini 3 Pro	Perplexity Sonar / Sonar Pro	Source
GPQA Diamond	Not published as Copilot-specific	91.9%	47.1% (Sonar)	Google Gemini 3 model card, Nov 18, 2025
MMLU / MMLU Pro	Not published as Copilot-specific	91.8%	68.9% (MMLU Pro, Sonar)	PricePerToken model profile, 2026
SimpleQA (F-score)	Not published	Not published	0.773 (Sonar) / 0.858 (Sonar Pro)	ithy.com comparison, updated Aug 16, 2026
SimpleQA (API accuracy)	Not published	Not published	92% at $0.052/request	Parallel.ai, Aug 19, 2026
Estimated monthly queries	~12 billion (Q1 2026)	Not disclosed separately from MAU	Not disclosed separately from MAU	AI Search Statistics 2026, aiseoshift.com
Monthly active users	~150M (“Copilot family”)	1 billion+ (Aug 11, 2026)	~100M (April 2026)	Google Blog / TechCrunch; RocketBlue; instantpress.co

Read the reasoning-benchmark gap carefully before drawing conclusions. Perplexity’s low GPQA and MMLU scores for its base Sonar model don’t mean Perplexity is “dumber” — Pro subscribers can route the exact same query through Gemini 3 Pro or GPT-5.x inside Perplexity’s own interface and get comparable reasoning performance, wrapped in Perplexity’s citation layer. The Sonar-specific scores only describe Perplexity’s default, cost-optimized fast path, not its ceiling.

Real-World Examples: How Teams Actually Use Each Tool

Specs only tell part of the story. Here’s how these three tools show up in actual workflows as of mid-2026.

• Finance analyst building a quarterly variance report: A financial analyst at a mid-size manufacturer uses Copilot directly inside Excel, referencing live formulas and pivot tables to draft variance commentary without exporting data to a separate chat window — the entire workflow stays inside the spreadsheet.
• Marketing team drafting a campaign brief: A marketing team on Google Workspace uses Gemini inside Docs to draft campaign briefs, then has Gemini auto-summarize the kickoff Meet recording directly into action items inside the same Doc, with no manual transcription step.
• Freelance journalist fact-checking a breaking story: A freelance reporter uses Perplexity Pro’s Deep Research mode to pull and cite primary sources on a fast-moving story, relying on inline citations to verify claims before publishing rather than trusting an uncited chatbot summary.
• Engineering team reviewing pull requests: A software team with a Microsoft 365 E5 license routes code review questions through Copilot Chat (running GPT-5.6) inside Teams, keeping the conversation attached to the same thread as the original pull request discussion.
• Small business owner without enterprise licenses: A solo consultant with no Microsoft 365 or Google Workspace Enterprise contract uses Perplexity Pro at $20/month for competitive research and market sizing, avoiding the $30–$87/user/month enterprise stack required by Copilot or Gemini Enterprise.
• University department on Workspace for Education: A university department already on Google Workspace for Education deploys Gemini’s education tier campus-wide, since it’s bundled into the existing Workspace agreement rather than requiring a new procurement cycle.
• Enterprise IT comparing total cost of ownership: An enterprise IT team modeling a 10,000-seat rollout finds Copilot Enterprise costs roughly $3.6 million a year in incremental spend on top of existing M365 E3/E5 licensing, prompting a side-by-side pilot against Gemini Enterprise’s lower $18–$25/user/month add-on before committing budget.

Best Use-Case Recommendations by Role

Which tool wins depends almost entirely on what suite you’re already paying for and what job you need done. Here’s the breakdown by role and scenario.

• Heavy Microsoft 365 users (Word, Excel, Outlook, Teams): Copilot wins on pure convenience — it’s already inside the apps you use daily, and GPT-5.6 is a genuinely strong model for drafting and analysis tasks once it’s embedded in your document.
• Google Workspace shops (Docs, Sheets, Gmail, Meet): Gemini is the obvious pick, especially given its 1M-token context window at the Pro tier and native Meet transcription/summarization that neither competitor matches as tightly.
• Researchers, journalists, and analysts who need sourced answers: Perplexity is purpose-built for this. Its citation-first design and Deep Research mode outperform both alternatives when the deliverable is “prove this claim,” not just “answer this question.”
• Budget-conscious individuals and students: Gemini’s free 32K-context tier or Perplexity’s free ~5 Pro Searches/day both beat paying for Copilot access outside a Microsoft 365 subscription, since Copilot’s real capability sits behind the Business/Enterprise paywall.
• Organizations without existing Microsoft or Google enterprise contracts: Perplexity Enterprise Pro at $40/user/month with zero prerequisite licensing is frequently the cheaper and faster-to-deploy option versus starting a new Microsoft 365 or Workspace Enterprise procurement from scratch.
• Teams needing model flexibility across vendors: Perplexity’s model picker (GPT-5.x, Claude, Gemini, Grok) is the only one of the three that lets a single subscription route across multiple foundation-model vendors instead of locking you into one.
• Multimodal-heavy workflows (video, image, long documents): Google AI Ultra’s Deep Think mode and native Imagen/Veo access make Gemini the strongest pick for teams doing serious image or video generation work alongside text.

Migration Guide: Switching Between Copilot, Gemini, and Perplexity

Switching your organization’s default AI assistant is rarely a clean cutover. Here’s a practical sequence for moving between any two of these three platforms.

1. Audit current usage patterns. Pull query logs or survey your team on what tasks the current tool handles — drafting, research, coding, spreadsheet work — before assuming the new tool covers the same ground.
2. Check your existing productivity-suite contract. If you’re locked into Microsoft 365 E3/E5 or Google Workspace Enterprise, switching away from Copilot or Gemini respectively means either paying for two AI layers simultaneously or waiting for contract renewal.
3. Run a 30-day parallel pilot. Give a small team (10–20 users) access to the target tool alongside their existing one rather than a hard cutover, so you can compare real output quality before committing budget.
4. Export and reformat saved prompts or Spaces. Perplexity Spaces, Gemini Gems, and Copilot’s saved prompt library don’t transfer automatically — budget time to manually recreate your team’s most-used prompts in the new tool.
5. Re-map integrations. Copilot’s value is tied to Microsoft 365 connectors; Gemini’s is tied to Workspace connectors; Perplexity’s is tied to its Comet browser and API. Confirm the new tool actually connects to your file storage, calendar, and communication stack before migrating fully.
6. Review data-handling and compliance terms. Enterprise tiers from all three vendors differ in what’s used for training and how long data is retained — get this in writing from procurement before moving sensitive workflows.
7. Train power users first. Identify the 3–5 employees who used the old tool most heavily and get them proficient in the new one first; they’ll become internal support for the wider rollout.
8. Set a hard cutover date and decommission the old license. Running two AI subscriptions indefinitely erases any cost savings from switching — pick a date, communicate it, and cancel the old seats on schedule.

Developer and API Access

For teams building on top of these platforms rather than just using the chat interface, API access differs sharply. Perplexity’s Sonar API is the most straightforward to call directly for search-grounded responses:

curl -s https://api.perplexity.ai/chat/completions \
  -H "Authorization: Bearer $PERPLEXITY_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "sonar-pro",
    "messages": [
      {"role": "user", "content": "Summarize the latest Gemini pricing tiers with sources."}
    ]
  }'

Gemini exposes its models through the standard Google AI API, with Google’s September 2026 lineup now spanning the budget-tier Gemini 3.8 Flash — $0.75 per 1M input tokens and $3.75 per 1M output tokens — up through the flagship Gemini 4 Argon at $2 per 1M input tokens and $10 per 1M output tokens, both available to developers at roughly 1M tokens of input context. Copilot, by contrast, doesn’t expose a direct consumer-facing chat API in the same way — developers instead build against the underlying GPT-5.6 family via OpenAI’s own API or through Microsoft’s Copilot Studio for enterprise-integrated agents, which is a meaningfully different developer experience than either Gemini’s or Perplexity’s self-serve API keys.

Pros and Cons of Each Platform

Microsoft Copilot

Pros: Deep, genuinely useful Microsoft 365 integration; runs OpenAI’s current flagship GPT-5.6 family; strong for document, spreadsheet, and presentation drafting tied to live enterprise data.

Cons: Requires an existing Microsoft 365 license, pushing effective enterprise cost to $66–$87/user/month; confusing, inconsistent public user-count disclosures; no officially documented context-window limit for everyday Copilot Chat use.

Google Gemini

Pros: Largest user base of any AI assistant at over 1 billion MAU; strongest published reasoning benchmarks (91.9% GPQA Diamond, 91.8% MMLU) among the three; generous free tier and clear, published pricing ladder from $0 to $250/month.

Cons: Free and mid-tier context windows (32K–128K tokens) are far below the marketed 1M-token ceiling reserved for Pro/Ultra; AI Ultra’s roughly $250/month price is steep for individual users; enterprise Gemini pricing is opaque and custom-negotiated rather than published.

Perplexity AI

Pros: No prerequisite productivity-suite license required at any tier; citation-first design is uniquely strong for research and fact-checking; Pro plan’s model picker gives access to multiple frontier vendors under one subscription.

Cons: Base Sonar model trails Gemini significantly on reasoning benchmarks (47.1% vs 91.9% GPQA Diamond); Comet browser adoption numbers are unreliable and contradictory across trackers; Enterprise Max’s $325/user/month is the steepest single-seat price among all three platforms.

Mobile Apps and Platform Availability

All three products now run across the full desktop-and-mobile spread, but the depth of that presence differs. Microsoft Copilot ships as a standalone app on iOS and Android, as a sidebar inside Edge, as a built-in panel in Windows 11, and inside every Microsoft 365 desktop and web app. That breadth is also why its user-count numbers are so hard to pin down — a single person touching Copilot in Word, then again in Windows, then again in Bing, can get counted as three separate “active users” depending on which surface a given tracker measures.

Gemini benefits from default placement on Android, where it has effectively replaced Google Assistant as the system-level voice assistant on most 2025-and-newer devices — a role reinforced in September 2026 when Google’s Gemini 3.8 Live Extended Thinking model posted a score of 82.6 on the Speech-to-Speech Quality Index — plus a standalone app on iOS and deep hooks into Chrome. That default-on-Android position is arguably the single biggest driver behind Gemini’s 1-billion-MAU milestone — it’s the AI assistant hundreds of millions of people never had to actively choose to install.

Perplexity is the only one of the three without a first-party operating-system placement advantage, and its growth to 100 million MAU has come almost entirely through deliberate installs — the mobile apps, the browser extension, and now the standalone Comet browser, which reached parity across macOS, Windows, Android, and iOS by March 2026. That’s a harder growth path than Gemini’s or Copilot’s, which makes Perplexity’s user-count trajectory arguably the most organic of the three, even though it’s the smallest in absolute terms.

Customer Support, Uptime, and Reliability

Reliability expectations differ by tier and vendor. Microsoft 365 Copilot inherits the same enterprise SLA structure as the rest of Microsoft 365 — Business and Enterprise customers get financially backed uptime commitments and access to Microsoft’s enterprise support channels, while free Bing/Windows Copilot users get community forums and no formal SLA. Google follows a similar split: Gemini for Workspace customers on Business Standard and above inherit Google Workspace’s enterprise support tier, while free Gemini app users rely on Google’s standard consumer help center.

Perplexity’s support structure is the least mature of the three by virtue of being the youngest, smaller company. Enterprise Pro and Enterprise Max customers get dedicated support and SSO-backed account management, but Perplexity doesn’t yet carry the decades of enterprise SLA precedent that Microsoft and Google bring from their existing productivity-suite businesses. For risk-averse enterprise buyers, that’s a real factor weighing in Copilot’s and Gemini’s favor even when Perplexity wins on price or feature fit — procurement teams tend to value a vendor’s support track record as much as the product itself.

Enterprise Considerations and Total Cost of Ownership

Total cost of ownership diverges sharply once you factor in prerequisite licensing. Copilot’s $30/user/month Enterprise add-on looks competitive against Perplexity’s $40/user/month Enterprise Pro until you remember Copilot requires an existing $36–$57/user/month Microsoft 365 E3/E5 license underneath it — pushing the real cost to $66–$87/user/month. Gemini Enterprise’s $18–$25/user/month add-on is cheaper on paper, but it similarly assumes a Workspace Enterprise contract is already in place.

Perplexity is the only one of the three that publishes a standalone enterprise price with no prerequisite license: $40/user/month for Enterprise Pro, scaling to $325/user/month for Enterprise Max’s heavier research and video-generation workloads. For an organization evaluating AI tooling from scratch — without an existing Microsoft or Google enterprise agreement — Perplexity’s flat, unbundled pricing is easier to model and often cheaper for teams under a few hundred seats. For organizations that already have thousands of Microsoft 365 or Workspace seats under contract, the marginal Copilot or Gemini add-on cost is usually the more defensible line item, since the base license is sunk cost either way.

Privacy, Data Handling, and Security Differences

Enterprise buyers consistently flag data handling as a deciding factor between these three. Microsoft 365 Copilot’s enterprise tiers explicitly exclude customer work data from model training, a commitment Microsoft has repeated across its Copilot pricing and compliance documentation. Google similarly offers enterprise-grade data controls under Gemini for Workspace, with data processing terms distinct from the free consumer Gemini app. Perplexity’s enterprise tiers (Pro and Max) include SSO and dedicated data-handling terms separate from its free and individual Pro plans, though because Perplexity routes queries through third-party foundation models (OpenAI, Anthropic, Google) depending on what the user selects, enterprise buyers should confirm exactly which downstream model providers see query data under their specific contract — a diligence step that’s simpler with Copilot or Gemini, where the model provider is fixed.

The Verdict: Which AI Chatbot Wins in 2026

There isn’t a single winner here, and the data makes that clear rather than ambiguous. Judged purely on scale and published reasoning benchmarks, Gemini wins outright: 1 billion-plus monthly active users, a 91.9% GPQA Diamond score that neither competitor’s public numbers approach, and a pricing ladder that scales from free to $250/month depending on need. If you’re picking one general-purpose AI assistant with no other constraints, Gemini’s combination of reach, benchmark performance, and transparent consumer pricing is the strongest overall package as of August 22, 2026.

But “wins outright” isn’t the same as “wins for you.” If your organization runs on Microsoft 365, Copilot’s GPT-5.6 backend embedded directly in Word, Excel, and Teams will save more time than a technically smarter model that requires switching apps. And if your job is research, journalism, or anything where an uncited claim is worthless, Perplexity’s citation-first design and multi-model picker beat both alternatives regardless of what a GPQA leaderboard says. Buy the suite you already live in, or buy the tool built for your specific job — don’t buy the biggest number on a benchmark chart.

Frequently Asked Questions

Is Microsoft Copilot the same as ChatGPT?

No. Copilot is Microsoft’s product wrapper around OpenAI’s models — as of July 9, 2026, Microsoft 365 Copilot’s preferred model is GPT-5.6, the same model family available directly through ChatGPT, but Copilot adds Microsoft 365 integration, enterprise data controls, and pricing that ChatGPT doesn’t offer.

Which is cheaper for a small business: Copilot, Gemini, or Perplexity?

Perplexity is usually cheaper for a small business without existing Microsoft or Google enterprise contracts, since its $40/user/month Enterprise Pro tier requires no prerequisite productivity license. Copilot and Gemini both assume you’re already paying for Microsoft 365 or Workspace, which adds $14–$57/user/month before the AI add-on even applies.

Does Perplexity use its own AI model or someone else’s?

Both. Perplexity built its own Sonar and Sonar Pro models for fast, cost-efficient, citation-backed search, but Pro and Enterprise subscribers can also route queries through third-party frontier models including GPT-5.x, Claude Sonnet/Opus 4.6, Gemini 3 Pro, and Grok via a model picker.

How big is Gemini’s context window compared to Copilot and Perplexity?

Gemini’s context window scales by subscription tier: 32K tokens free, 128K tokens on AI Plus ($4.99/month), and up to 1M tokens on AI Pro ($19.99/month) and AI Ultra (~$250/month). Copilot’s context window isn’t officially published for everyday Copilot Chat use, with independent estimates around 128K tokens and observed practical limits closer to 25,000–40,000 tokens. Perplexity likewise doesn’t publish an official context-window figure for its Sonar models.

Why do Microsoft Copilot’s user numbers vary so much between sources?

Microsoft’s own disclosures put “Copilot family” monthly active users at roughly 150 million, but that figure spans multiple surfaces — Microsoft 365 Copilot, Windows Copilot, Bing Chat, and GitHub Copilot — that other trackers count separately or combine differently. Independent trackers report figures as low as 33 million (narrow app-only definitions) and as high as 420 million (every surface combined), which is why analysts treat any single Copilot MAU figure with caution.

Can I use Perplexity, Gemini, and Copilot together instead of picking one?

Yes, and many teams do. A common pattern is Copilot or Gemini for day-to-day document and email work tied to whichever productivity suite the organization already pays for, with Perplexity layered on top specifically for research tasks that need citations. The main cost to weigh is running multiple subscriptions rather than any technical incompatibility.

What is Perplexity’s Comet browser and do I need it?

Comet is Perplexity’s dedicated AI-native browser, free since its global rollout completed across macOS, Windows, Android, and iOS by March 2026. It’s not required to use Perplexity — the core chat and search product works fine in any browser or the mobile app — but Comet adds tighter agentic browsing features for users who want Perplexity embedded at the browser level rather than in a separate tab.

Does Copilot Enterprise or Gemini Enterprise train on my company’s data?

Both Microsoft and Google state that enterprise-tier data is excluded from model training under their respective enterprise agreements, distinct from free consumer-tier usage. Enterprise buyers should still confirm the exact data-retention and training-exclusion terms in their specific contract before deploying either tool on sensitive data.