Dots vs Grok Bot vs Muse, One Month Later: What Actually Shipped
One month after three always-on agents collided into the market in the same three weeks, the gap between them has widened — in genuinely different directions.
Three always-on agents launched within the same three-week window last September: OpenAI's dots on September 29, Meta's Muse on September 8, and xAI's Grok Bot, already in beta since August and extended to businesses in early September. The original side-by-side on their permission models compared them at launch. A month into dots' public life, the three products have moved in genuinely different directions, and the comparison worth making now is about what actually shipped, not what each company promised.
OpenAI: incremental, broad, and quiet about dots specifically
OpenAI's confirmed October changelog is dense but almost entirely dots-adjacent rather than dots-specific: Locked Chats rolling out everywhere on October 20, Finances in ChatGPT expanding to Free and Go on October 2, mobile PDF scanning and virtual try-on on October 1, GPT-Live-1.1 voice options on October 15, and a cheaper sibling model, GPT-6.1 Sol, announced alongside dots at launch but not confirmed to run them. None of this resolves the real open questions — additional-dot pricing, what happens as the usage grace period ends, whether third-party plugins work yet, or when the EEA/UK/Switzerland exclusion lifts. OpenAI shipped broadly and said little about dots directly.
xAI: the riskiest, most consequential move of the month
Grok Bot took the most structurally significant action of any of the three products this month, and it has nothing to do with features in the usual sense. On October 8-9, Elon Musk confirmed on X that Grok Bots now run Grokipedia's entire content pipeline — creation, fact-checking, edit review, and curation — as a coordinated multi-agent workflow, replacing the earlier single-model process. Grokipedia has no volunteer-editor community and no talk-page consensus process the way Wikipedia does; this move removes the last layer of human editorial review from an encyclopedia that, per secondary reporting attributed to Wikipedia/Grokipedia's own figures, had grown from 885,279 articles at its October 2025 launch to more than 6 million by early 2026. That's the real story of Grok Bot's month — not the smaller, officially confirmed changes to its own product (Secure Forms and a reconnect prompt on October 7, Team Bots removed from the primary picker on October 5, both per xAI's own changelog), but a decision about what an AI agent is now trusted to run unsupervised at encyclopedic scale.
Meta: a real model upgrade, undercut by its first public growth-slowdown report
Meta shipped the most substantively new capability of the month: Muse Wonder and Muse Wonder Full, released October 12, 2026 and described by Meta as its first frontier model. Meta's own model card claims benchmark parity with Fable 5.1 and GPT-6 Astra, and a lead over Fable, Opus, and Astra specifically on MemoryBench. Independently, the Artificial Analysis Intelligence Index scores Wonder Full at 56 and Wonder at 52 — the more load-bearing numbers, since they come from a third party rather than Meta's own materials. Wonder is available now via the Meta Model API and Muse Code at $4 per million input tokens and $20 per million output tokens; Wonder Full is Muse Code-subscribers-only and not served via the API. Neither model ships with open weights, a real departure from Meta's Llama-era posture that the announcement doesn't explain.
That capability story sits next to Meta's first real controversy. Hunterbrook Media/TickerTrends, in a report published October 9, found Muse downloads falling to roughly 170,000 on October 6, down 21.1% from a roughly 215,000 peak on September 24, with weekly DAU growth slowing 62.4% from its early surge even as absolute DAUs hit a series-high 2,730,919 on October 6. Hunterbrook Capital was short Meta stock at the time of publication — a disclosure that has to accompany any citation of these numbers. A separate firm, Sensor Tower, reported via Forbes that Muse hit 5 million-plus US downloads in its first 22 days — a different, cumulative, US-only metric that doesn't actually contradict TickerTrends' daily-rate figures, just measures something else.
One month in: OpenAI shipped broadly and said little about dots directly. xAI made the single highest-stakes decision of the month, handing an AI agent the entirety of an encyclopedia's editorial process. Meta shipped a genuine capability upgrade while absorbing its first public growth-slowdown report, from a source with a disclosed financial interest in the story being bad.
Reading the growth numbers against each other, carefully
It's tempting to line Muse's download slowdown up against Grok Bot's reported DAU surge — from roughly 155,000 on September 20 to roughly 450,000 on October 17, per an X trending-page summary rather than an xAI press release — and declare a winner. That would be a mistake on both sides. The Grok Bot figures come from a single secondary source with no disclosed methodology checked here, attributing the growth to faster task handling, voice calls, email aliases, and X DM support as new surfaces; treat the specific numbers as an unverified claim, not an xAI-confirmed one. The Muse figures, while independently reported by two different firms, measure different things from each other — TickerTrends' daily download rate and app-review volume on one hand, Sensor Tower's cumulative 22-day US download total on the other — and neither is directly comparable to Grok Bot's DAU claim, which is yet another different metric entirely. None of these three numbers should be placed on the same chart and compared as if they measured the same thing.
Why xAI's move is the one worth the most scrutiny
Of the three stories this month, Grokipedia's handoff to Grok Bot deserves more scrutiny than its framing in most coverage has given it. A secondary source frames the move as "a feature or a risk" without picking a side, which is a reasonable analytical stance — but it also means nobody has yet published a clear account of how the new multi-agent workflow handles contested claims, rapidly changing information, or disagreement between the agents performing creation versus fact-checking versus curation in the same pipeline. As of February 2026, per earlier reporting, Grok was already handling more than 75% of edit suggestions and approvals itself, with human involvement limited to submitting suggestions through a form rather than editing directly. Removing the remaining layer of human review doesn't change that structure so much as remove the last checkpoint sitting on top of it — a meaningfully different kind of change than a feature update, and one none of the three companies' other October announcements come close to matching in scope.
What hasn't changed: nobody has published a real safety report
Despite a month of real activity, none of the three companies has published a full safety or red-team report for its newest high-stakes capability — not OpenAI's post-grace-period dots terms, not xAI's Grokipedia editorial safeguards, not Meta's promised-but-not-yet-published Muse Wonder safety report. That gap, flagged at the close of the original launch series, is the one constant across all three products a month later.
Part 46 of 47 in the ChatGPT Dots series. Previously: Enterprise Dots, Still Off by Default: Who Is Actually Testing Them. Next: The First Month of Dots: A Scorecard Against What OpenAI Promised. Browse more builder insights or explore AI skills at aiskill.market.