ChatGPT · Gemini · Claude — Jan 2025 – Oct 2026 | iOS + Android | All Countries
Source: Sensor Tower
Updated: Oct 2026
Monthly Trend
DAU / WAU / MAU / Downloads over time
DAU / MAU Ratio
User stickiness — Daily Active as % of Monthly Active
iOS vs Android
Platform breakdown
Market Share Evolution
% share among 3 apps
Avg Time Spent
Mobile app engagement per active user
Top 20 Countries
Oct 2026
Country MAU Growth
Sep → Oct 2026 · MAU Net Change & MoM %
Country
ChatGPT
Gemini
Claude
Apr MAU
Net
MoM
Apr MAU
Net
MoM
Apr MAU
Net
MoM
Monthly Net Increase
Net new users per month (M)
New User Share Distribution
% of monthly net increase captured by each app
MoM Growth Rate
Month-over-month growth %
Ratio vs ChatGPT
Gemini & Claude as % of ChatGPT
AI App Web Intelligence
ChatGPT · Gemini · Claude — Jan 2025 – Oct 2026 | Web Traffic | Global
Source: Semrush
ChatGPT / Claude / Gemini: Oct 2026
Monthly Trend
DAU / WAU / MAU over time
DAU / MAU Ratio
User stickiness — Daily Active as % of Monthly Active
Market Share Evolution
% share among 3 apps
Avg Visit Duration
Average time per visit (Semrush)
Top 20 Countries
Sep 2026 · Unique Visitors
Country MAU Growth
Jul → Sep 2026 · Unique Visitors Net Change (K) & % Change (Aug skipped — no country-level export for that month)
Country
ChatGPT
Gemini
Claude
Jul MAU
Net
%Chg
Jul MAU
Net
%Chg
Jul MAU
Net
%Chg
Monthly Net Increase
Net new users per month (M)
New User Share Distribution
% of monthly net increase captured by each app
MoM Growth Rate
Month-over-month growth %
Ratio vs ChatGPT
Gemini & Claude as % of ChatGPT
Coding Agent & B2B Adoption
Claude Code · Codex · npm · GitHub · Ramp AI Index
Source: npm + GitHub + Ramp
Updated: Oct 2026
Public GitHub Repository Activity
OpenAI vs Anthropic · Jan 2025 – Nov 2026 · Official public repository activity · Collected Nov 9, 2026
OpenAI · October 2026
3,062,179
-29.0% MoM
4,073
+2.3% MoM
Anthropic · October 2026
2,057,756
-86.4% MoM*
2,483
-19.0% MoM
*Anthropic's October drop looks dramatic only because September's base was inflated by a single bulk import (see the anomaly note below) — excluding that one commit, September was ~1.32M and October's 2.06M is actually a modest increase. November (MTD thru Nov 9): OpenAI 1,002 commits / 894,643 lines; Anthropic 614 commits / 309,296 lines.
Monthly Lines Changed
Text additions + deletions · Merge commits excluded
Monthly Commits
Monthly commit count · Merge commits included
Methodology:321 public, non-fork repositories. History reachable from each current default branch, grouped by UTC committer month. Commit counts include merges; lines changed sum additions and deletions, exclude merges, and detect renames. Documentation, generated files and datasets are included. This measures organization repository activity; the Claude Code co-authored commits below measure tool usage.
Methodology notes and interpretation
Historical attribution:Anthropic’s OpenROAD mirrorwas created in August but inherited upstream history. It contributes 353,465 lines to June, or 40.8% of Anthropic’s monthly total. GitHub does not mark it as a fork, so it is included.
Bulk file changes:One ten-proofs commitcontributes 550,770 lines, mainly Lean proof files. tunnel-client also includes substantial additions and deletions of SPDX software component inventories.
Commit date versus publication date:Anthropic’s commerce-agents repository was created on September 1. A commit dated August 31 contributes 88,497 lines to August.
Bulk import (September):anthropics/fermats-last-theorem added 13,841,172 lines across 62,591 files in one shot — an initial import of a large collaborative Lean formalization project, not incremental development. It accounts for ~91% of Anthropic's September lines-changed total; excluding it, September would read ~1.32M, in line with other recent months.
October (new, November refresh):no comparable single-commit anomaly — October's largest individual commit (anthropics/formal-math, Oct 26) changed 153,191 lines, two orders of magnitude smaller than September's import. October's totals read as ordinary repository activity.
Interpretation:Public changes reflect repository activity, not a direct measure of engineering productivity, model progress or product adoption. Inherited mirror history and bulk imports require separate interpretation.
npm weekly download counts | ⚠ Codex weeks of Apr 27 & May 4 2026 (130M / 89M) are a CI/bot anomaly: ~181M of May's 239M came from May 1–5 (peak 46M/day vs 1.4M median) — not user adoption
Market Share Evolution
% share among CC + Codex by month
CC / Codex Ratio
Monthly download ratio (x times)
Cumulative Downloads
Running total since launch (weekly sampled) | ⚠ Codex total includes ~220M bot-driven downloads from the Apr 27–May 5 2026 anomaly
GitHub Commits (Claude Code)
Monthly co-authored commits, median of repeated GitHub Search queries | Codex does not tag co-author — not comparable
Weekly Active Users — Claude Code vs Codex
CC: commits ÷ 3.25 (anchor: Feb 5.2M commits = 1.6M WAU), thru Oct-26 | Codex: official disclosures thru Jun-26; Jul/Aug/Oct (dashed/hollow) use a combined Codex+ChatGPT Work tally as a proxy — not directly comparable to the Codex-alone points before them | Sep-26 has no Codex reading (no source found) | Oct-26 Codex figure (40M) is a third-party estimate (Every/Dan Shipper), not an OpenAI/Altman disclosure like the earlier points
Claude Opus vs GPT-6 — GitHub commit-message mentions (SAMPLE)
Weekly count of commits whose message contains the exact phrase, via GitHub's public Search API | 4 weeks, pulled live 2026-11-02
GPT-6 minus Claude Opus | opposite sign from the reference chart
Methodology:GET /search/commits?q="<term>" in:message committer-date:START..END, one query per week, exact-phrase match against commit message text only (not trailers/metadata). Unauthenticated — this session's authenticated token came back "flagged as spammy" by GitHub's abuse system, so these counts are from anonymous requests, capped at 10 req/min with intermittent 403s requiring retries.
Why this is a sample, not a tracked metric: every "Claude Opus" query returned incomplete_results: true — GitHub's Search API explicitly didn't finish scanning within its time budget, so these are a partial, reproducibility-uncertain lower bound, not a full count (same caveat this file already carries for the "GitHub Commits (Claude Code)" chart above, which uses the same API). A naive "claude opus" query without in:message returned total_count = 9,156,739 for a single week — almost certainly the API's own broad-match fallback, not a real count; narrowing to in:message is what brought it down to a plausible ~2,000/week. The resulting numbers also don't resemble the reference chart's reported percentages at all (this pull: Opus flat, GPT-6 up sharply; reference: Opus up sharply, GPT-6 down sharply) — TickerTrends almost certainly uses a more complete source (e.g. the full GH Archive via BigQuery, which indexes every public commit event rather than relying on GitHub's best-effort live Search API) and/or a different normalization (share of total weekly commits, not raw counts). Reproducing their exact figures would need that same comprehensive dataset, not this API.
VS Code Daily Installs
30-day moving average (revealera.com's own field name — this page previously mislabeled it 84-day; corrected back after a 2026-11-09 recheck against the live API, which explicitly states "30-day") | thru 2026-11-07 | Claude Code up ~9% since Oct 18 (74.8K→81.7K/day) after an Aug-Oct decline, Codex up ~10% since Oct 18 (38.6K→42.4K/day) | Source: Bloomberry.com
B2B AI Subscription Penetration — US businesses with paid subscriptions, Ramp AI Index
Sep 2026: Anthropic 43.8% leads OpenAI 40.6% · Full series refreshed 2026-10-20 · source ↗
All series pulled live from ramp.com/data/ai-index (fetched 2026-10-20). Per-vendor shares overlap (multi-vendor customers); Ramp revises history, so refresh the full series, not just the latest month. Latest available month: September 2026. Google was removed from Ramp's own vendor breakdown as of this pull and is no longer tracked here.
Net add vs Aug-26: +$4.7B (from $69.6B) · YipitData's own preliminary estimate thru 9/27, subject to revision once the B2B panel's full September data lands (~2 weeks after this file)
OpenAI ARR (Sep-26, Prelim.)
$47.8B
Net add vs Aug-26: +$6.8B (from $41.0B) · closely aligns with the FT's own "approaching $50bn" reporting
Update note (Oct-08 data files): extended thru Sep-26 on the ARR/subscriptions/net-adds series (consumer e-receipt panel has only a ~2-day lag); B2B-panel-dependent series (B2B spend, adoption, big-account count, channel mix) extended only thru Aug-26, since the B2B panel itself won't have full September data for another ~2 weeks per YipitData's own note. Both Sep-26 ARR figures are explicitly flagged "Preliminary" by YipitData and may move once that panel data lands. One real restatement this round: OpenAI's Jul-26 disclosed anchor was rebased from $38.0B to $32.0B — YipitData explicitly cites a more precise FT report superseding the earlier, vaguer Bloomberg "more than $40 billion" phrasing used to set the prior figure. Anthropic's Jul-26 anchor ($65.0B) is unchanged. This page's own self-derived ARR share (the chart below) puts OpenAI at 39% in Sep-26, up from 37% in Aug-26; YipitData's own narrative "ARR Share" line (based on the disclosed/interpolated anchor, a different basis than this chart's continuous antYipit/oaYipit series) tells a similar directional story but a different level — 39% in Sep-26 vs. 32% in Jun-26, a 3rd straight month of OpenAI gains by their own count. Anthropic's B2B panel adoption continued widening its lead over OpenAI's (80.4% vs 79.0% as of Aug-26).
ARR Growth Trajectory
Dashed = company-disclosed only (months YipitData interpolates are left blank) · Solid = YipitData estimate · $B · both series frm Jan-25 · latest point Sep-26, YipitData's preliminary estimate thru 9/27
Monthly Net-New ARR (total run-rate revenue)
M/M change in YipitData run-rate estimate · $B · Jan → Sep 2026
B2B Daily Spend Growth (raw panel data)
B2B Spend Panel · average spend per day ($K/day) · log scale · Jun 2023 – Aug 2026 · panel runs ~2 weeks behind, so Sep-26 is not published yet
Panel-observed active subs (EOM) · Claude Pro+Max vs. ChatGPT Plus+Pro+Go · Nov 2024 – Sep 2026
Subscriptions Net Adds
M/M change in panel-observed subs · through Sep-26 · Claude's monthly add bottomed in Aug-26 before ticking back up in Sep-26, while ChatGPT's has now accelerated three straight months to the strongest add in this entire series
B2B Panel Adoption
Share of B2B spend panel with any observed spend on each vendor · Jun 2023 – Aug 2026 (latest B2B month) · Anthropic has now led for two straight months (80.4 vs 79.0 in Aug-26, up from 79.8 vs 79.0 in Jul-26)
Big-Account Count (>$100K annualized)
B2B Spend Panel · Jun 2023 – Aug 2026 (latest B2B month) · Anthropic 119 accounts >$100K vs OpenAI 50 in Aug-26
Monthly Net-New Adopting Companies (NEW)
Same 674-company Yipit B2B panel sample, OpenAI vs Anthropic · Jul 2025 – Jun 2026 · not refreshed by the Oct-08 file (no counterpart sheet) — still on the Aug-05 basis
Adoption MoM Growth Rate (NEW)
M/M % growth in adopting-company count, same sample · Jul 2025 – Jun 2026 · not refreshed by the Oct-08 file — Anthropic's growth rate cooled from its Mar-26 peak (13%) to under 1% by Jun-26 on that older basis
PROPOSED NEW SECTION — sample for review, not live data
Personal AI Agents · Head-to-Head
Musevs. Grok Bot
Both now on real Sensor Tower tracking — Muse on App Store (iOS), Grok Bot on a full Unified (iOS+Android, 81 countries) export. Meta's ad-subsidized giant against xAI's pay-per-use agent: same category, opposite business model, and — now that both have real multi-week data — a much closer growth race than the single-point press comparisons ever showed.
Source: Sensor Tower (Unified + App Store)
As of: Nov 7, 2026
Meta
Muse
2.06M
▲ new high, still climbing
4.13M
+7.5% vs prior wk
8.51M
vs 7.48M in Oct
9.83M/60 days
now 1.33× Grok Bot's cumulative
$722K/Sep 9–Nov 7
1/6.5th of Grok Bot's cumulative
15.3min, latest
~1.8× Grok Bot's
Free + ad-subsidized. D14 retention 26.9% on a 72.9% D1 — the structural boost from mandatory card-on-file and async task check-ins. App Store (iOS) data only — Sensor Tower has no Android coverage for Muse at all, not just an unpulled export.
vs
xAI
Grok Bot
628K
plateaued ~620–660K since mid-Oct
1.35M
flat since 10/26 (1.37M)
2.38M
still climbing: 142K(Aug)→2.38M(Nov)
~7.4M/86 days
now trails Muse's 9.83M, despite its 4-wk head start
$4.71M/Aug 11–Nov 4
6.5× Muse's, since each launch
8.6min, latest
down from ~10–11 min in Aug
No free tier — paid-only, metered by token. Now on real Sensor Tower Unified data (iOS+Android, 81 countries), not press estimates. D14 retention 6.1% off a 23.3% D1 (download-weighted blend of its separate iOS/Android retention panels — Sensor Tower doesn't pre-combine retention across platforms the way it does DAU/downloads). Nov 1: Musk calls serving cost "unsustainably high relative to revenue."
Daily Active Usersreal Sensor Tower data for both, Aug 11 – Nov 4
Grok Bot launched 4 weeks before Muse and started two orders of magnitude smaller — by November the two are within ~3.3× of each other
Grok Bot's DAU climbed almost 70× from its Aug 11 launch (8.9K) to an October plateau (~620–660K) and has stayed there since. Muse took a different path: it also leveled off in October around 1.7M, but broke back out in late October and kept climbing into November, hitting a new high of 2.06M on Nov 6 — the two apps are no longer following the same flat-since-October shape.
Weekly Active Usersreal data, both apps
Aug 10 – Nov 2, weekly · real data for both apps across the full window
D1 → D14 RetentionSensor Tower, Sep cohort
Full D1–D30 decay curve, both apps, same Sep-2026 install cohort · Grok Bot blended from its separate iOS/Android panels (26% iOS / 74% Android, by Sep download share)
With the full curve now pulled directly from Sensor Tower (not press screenshots), the story changes in one important way: Muse opens far hotter (72.9% D1 vs Grok Bot's 23.3%) and actually holds its D1 cohort better over the first two weeks — 36.9% of Muse's D1 users are still there on D14, vs 26.4% for Grok Bot. (An earlier version of this page, built from a secondhand press chart, had this reversed — this direct pull supersedes it.) By D30 both curves have mostly flattened: Muse 20.3%, Grok Bot 5.6% — Muse's absolute lead holds all the way out, even as both lines decay most steeply in the first week.
User Stickiness — DAU ÷ MAUmonthly, both apps
Average daily actives that month, divided by that month's monthly actives · higher = a typical monthly user opens the app more days per month · November is month-to-date, both sides
Counterintuitive given everything above: Grok Bot is the stickier product on this metric, not Muse — 26.8% DAU/MAU in October vs Muse's 19.5%, and the gap has been consistent since each app's early weeks. A DAU/MAU of ~27% means a typical Grok Bot monthly user opens it roughly 8 days a month; Muse's ~20% is closer to 6 days. This doesn't contradict Muse's higher time-spent-per-day or better D14 retention — it suggests Muse's much larger monthly base includes a wider tail of casual, occasional users (consistent with a free, ad-driven funnel), while Grok Bot's smaller, paid-only base skews toward people who committed to using it regularly enough to justify paying for it.
Downloadsreal daily data, both apps
Aug 11 – Nov 4 · Grok Bot: Unified (iOS+Android). Muse: App Store (iOS) only
Grok Bot launched a month earlier and built an early lead, but Muse has now pulled ahead on cumulative downloads since its own launch — 9.83M for Muse (60 days) vs 7.36M for Grok Bot (86 days) — despite Grok Bot's 26-day head start.
Revenuereal daily data, both apps
Aug 11 – Nov 4 · In-app net revenue, same Sensor Tower pipeline as Downloads
Revenue is still where the two apps genuinely diverge: Grok Bot's cumulative revenue since launch ($4.71M) is ~6.5× Muse's ($722K) — even though Muse now has more downloads and a bigger active-user base. Paid-only beats free-and-ad-subsidized on revenue, even while trailing (now) on reach.
Engagement — Avg. Time Spent per Dayreal daily data, both apps
Minutes per day per active user, session-count-weighted across all 81 tracked countries
Muse users spend roughly 15.3 minutes/day in the app, about 1.8× Grok Bot's 8.6 minutes/day (which has also been declining since Grok Bot's August launch spike of ~10–11 min). Muse's edge in time-on-app is consistent with it also winning on D1 retention and absolute reach — it's a bigger part of more users' daily routine, even if Grok Bot's users convert to revenue more effectively.
Who's Actually Using These Apps — Age DistributionSensor Tower demographics, US
% of users by age bracket (men + women combined) · Muse: Jun–Oct 2026 window · Grok Bot: Nov 2025–Oct 2026 window
Neither app is a youth product: 18–24 is the smallest bracket for both (7.2% Muse, 8.8% Grok Bot). The single largest segment for both is 35–44 (33.3% Muse, 29.2% Grok Bot), and the broader 25–54 working-age band covers roughly 70% of users on both sides. 55+ users are a meaningful ~19–21% of the base too — not a rounding error. The honest read: adoption is already broad-based across working-age adults, not concentrated in one niche age group. That breadth is a better foundation for durable growth than a youth-only spike would be — but this is a snapshot of who uses it today, not proof of a rising trend by itself. The actual evidence for "usage is still climbing" is the DAU/MAU charts above, where both apps kept adding users into November rather than flattening.
Launch Order
Grok Bot beta preceded Muse's public launch by 4 weeks
xAI
Grok Bot (beta)
Aug 11, 2026
Meta
Muse (US)
Sep 8, 2026
Business Model & Pricing
Why the revenue numbers above look the way they do
Product
Model
Free Tier
Paid Tiers
Cost Strategy
Muse
Not disclosed; own Secure-VM-per-user sandbox
Yes — 100M tokens/week (card required)
Power $20/mo (500M/wk) · Maximum $100/mo (3,000M/wk)
Subsidized by Meta's $250B+ ad business — in-house compute costs ~1/7 of API market price
Grok Bot
Grok; hosted VM per bot
None — Cursor or SuperGrok subscription required
Overage billed at $2/M input · $6/M output tokens, no stated cap
Costs passed directly to the user — no in-house inference subsidy disclosed
The pattern holds even more clearly now with real revenue data: Muse can give away 100M tokens a week because Meta's ad business backstops the loss, while Grok Bot has no equivalent subsidy — no free tier, direct overage billing — and it's the one of the two whose own founder has said the serving economics don't work yet. And yet it's Grok Bot's paid-only model, not Muse's free-tier reach, that's generating the larger revenue number.
SemiAnalysis · Anthropic IPO Financials
"Anthropic 3Q26 Profit Over $1B: The Anthropic IPO Financials Sneak Peak" | Tokenomics Model — bottom-up by SKU / tier / customer | Confidential IPO filing Jun 1, 2026
20× 2027E ending ARR of $300B — would be the world's largest company
Key Takeaways
AI labs can be wildly profitable. Anthropic turns GAAP-profitable in 3Q26E ($1.2B EBIT, 6% margin) while growing NNARR faster than OpenAI, which sits at roughly -100% EBIT margins.
The business model is the moat: 75–85% of revenue is usage-based API (subscriptions just 15%, consumer only 5%), driving blended gross margin from -94% in 2024 to ~65% in 2026 (API GM 80%+).
Reinvestment flywheel: 2027E revenue-after-COGS of $160B vs OpenAI's $92B, and a cumulative EBTIT advantage of ~$250B through 2028, funds the next SOTA model and widens the lead.
Valuation: base case 20× 2027E ending ARR of $300B = $6T enterprise value; requires NNARR accelerating to ~$15B/month in 2027 (vs $10B+/mo today).
Quarterly P&L: Revenue → GAAP Profitability
Consolidated income statement, $M · 3Q26/4Q26 = SemiAnalysis estimates · EBTIT = earnings before training, interest, tax
Cumulative training spend since 1Q24: $8B → yielded a $60B ARR business · >60% of compute is training/internal today
Customer Momentum
Customers >$100k/yr: +7× in past year · >$1M/yr: +42× in 2 years · avg Claude Code enterprise user just $150–250/mo
Reinvestment Firepower
2027E revenue-after-COGS: $160B vs OpenAI $92B · long-run: GM mid-70s, training 48% of compute, opex 20% of rev
Steady-State Margins
30–40% EBIT/FCF margins possible at maturity — "like the best software companies of yesteryears"
Risks (SemiAnalysis watchlist)
Tokenbudgeting — enterprises capping spend; but Global 2000 AI spend is still tiny vs IT budgets, 90% of Claude Code users spend <$30/day.
OSS & price cuts — OpenAI rumored to cut token pricing to win back spend.
Google / Meta coding models — a 4-horse frontier race would pressure token pricing and gross margins.
Regulatory model blockades — Fable-style government delays at the frontier.
Compute shortage — where will the supply come from?
Upside
Fable — higher pricing mix per M tokens + new workloads accelerate NNARR above the current $10B/mo pace in 2H26.
Cyber — Mythos/Fable as the next major vertical, ramping faster than coding did.
Seat ramp — Global 2000 / Fortune 500 deployment still nascent; per-seat budgets rise with ROI.
New s-curves — healthcare, finance, biotech verticals stack on top; $300B end-2027 ARR "easily possible" at $15B/mo NNARR.
AI Capacity Tracker
OpenAI vs. Anthropic — status-page incidents as a compute-pressure proxy | Official status-page snapshots · incident archive
Source: incident.io + Atlassian Statuspage
Archive: –
Refreshed: 2026-11-09
OpenAI — Capacity Events (90d)
–
–
Anthropic — Capacity Events (90d)
–
–
Key Takeaways
Refreshed November 9, 2026: the incident archive contains 259 OpenAI and 574 Anthropic records. Using the existing keyword classification, the trailing 90-day archive contains 55 OpenAI and 37 Anthropic capacity-related events — OpenAI still leads the 90-day total, as it has since the October 20 snapshot (61 OpenAI vs. 46 Anthropic) reversed Anthropic's September 21 lead (58 vs. 75). Major/critical severity hasn't flipped the same way: Anthropic still accounts for most of it, 22 vs. OpenAI's 3 (was 23 vs. 3 on October 27).
October closed at 19 OpenAI vs. 15 Anthropic capacity-related events, down from September's 26 OpenAI vs. 9 Anthropic; November is month-to-date (thru Nov 9) at 3 OpenAI and 1 Anthropic so far. The current week is also partial; compare complete periods separately.
Coverage limitation: these are observed archive counts. Source retention and gaps between snapshots may omit incidents, so a lower count does not establish lower compute pressure. Earlier manually entered monthly totals have been replaced with counts reproducible from the incident archive. Cross-vendor absolute counts are not comparable.
Weekly Capacity Events — Trailing 18 Weeks
Side-by-side weekly count · current week is partial · absolute levels not cross-comparable (Anthropic reports per-model, OpenAI tends to consolidate) — compare each company's own trend
Monthly Capacity Events — Archived History
Anthropic back to Jul-25 (Atlassian Statuspage, ~12mo retention) · OpenAI only from Mar-26 (incident.io exposes ~3mo; earlier history is unrecoverable, this archive is now the only record going forward)
Recent Major / Critical Incidents
Last 30 days · sorted newest first · duration where disclosed