Personal Agent Bench

Corrections

When something we published turns out to be wrong or stated too broadly, we fix the page and log the fix with its date. Below are the corrections that changed a result or what we said about a product. Every other fix is counted by page further down, and the full log has each one's old claim, new claim and reason.

Exam rule changes are logged separately on what changed in the exam. To report an error, use the corrections line.

Each month's corrections are also counted and explained in the monthly report: October 2026 report.

Results and product facts

Dispatch and Cowork coverage now includes Ultra

We said: The page described Dispatch as in limited beta for Pro and Max plans, and the October cloud move as applying to Pro and Max.

We now say: Anthropic's help articles now say Dispatch is in limited beta on some Pro, Max, and Ultra plans, that the 6 October cloud move applies to Pro, Max and Ultra, and that the merged chat-and-Cowork experience covers Ultra with Team in beta and Enterprise where enabled. The pricing paragraph now lists Ultra at $200 a month.

support.claude.com articles 13947068 and 15520349, read 5 November 2026. Anthropic added the Ultra plan to both articles since the 22 October check.

Meta has now given a Muse usage figure

We said: We said daily active user figures for Muse came only from third-party estimates and that Meta had not published one.

We now say: Meta has given its own figures: on its 21 October earnings call Mark Zuckerberg said more than 10 million people in the US and Canada had set up Muse accounts and about 25% use it on any given day. We quote it as Meta's number and keep it apart from Sensor Tower's download estimate, which is a different measure.

Associated Press report of Meta's Q3 2026 earnings call, 21 October 2026 (syndicated at ksat.com), read 27 October 2026; TechCrunch, 20 October 2026, for the Sensor Tower estimate.

Muse is available in Japan, not only the US

We said: Muse was described as rolling out in the US since 8 September, with no other country named.

We now say: Muse is rolling out in the US since 8 September and, per Meta's Japanese newsroom, in Japan since 23 October, its first market outside North America; the Muse page has a new FAQ on what Meta says is available there.

Meta, about.fb.com/ja/news/2026/10/muse-personal-ai-agent-launches-in-japan/, published 23 October 2026 (JST), read 26 October 2026. The English newsroom feed did not carry it, so our news watch missed it; the Japanese newsroom is now a watched source.

Muse Spark now has an open-weight version

We said: Neither Muse Spark nor Muse Wonder had public weights; a 'muse spark open source' headline was said to be almost certainly about Glimmer or Muse Gadgets, not Spark itself.

We now say: On 15 October 2026 Meta released Muse Spark Open, which it calls the open-weight version of Muse Spark (816B total, 42B active), under its Muse Model License, which carries attribution and large-provider conditions. Meta does not say it is the model the Muse agent runs on. Muse Wonder still has no public weights. Alexandr Wang had said in August that Spark 1.2 would get an open-weight version.

Meta's post 'Introducing Muse Spark Open' (research.meta.ai/blog/muse-spark-open) and the Hugging Face page meta-models/Muse-Spark-Open-816B-A42B, read on 22 October 2026; Constellation Research, 10 August 2026.
2026-10-22/grok-bot/

Grok Bot access moved into The X Subscription

We said: Access comes with any paid individual Cursor plan (Pro is $20 a month); Cursor's pricing lists Pro $20, Pro+ $60, Ultra $200 and Teams $40 per user.

We now say: Cursor's pricing page now sells The X Subscription (Free $0, Plus $8, Premium $30, Super $100, Ultra $200, Teams from $30 per user) with Grok Bot from Premium up; Pro, Pro+ and Ultra are listed as legacy Cursor plans that still include it. The cheapest new route is $30, not $20.

cursor.com/pricing and cursor.com/help/grok-bot/plans, read on 22 October 2026.
2026-10-12/agents/town/

Town's per-assistant allow-lists: 29, not 33

We said: Town is the only product on this bench that applies per-agent allow-lists thirty-three times over.

We now say: Town applies per-agent allow-lists to each of the 29 assistant records that carry their own instructions.

The same page says four of the 33 records are shells without their own instructions and allow-lists.
2026-10-12/muse-vs-pine/

Muse and Pine completion rates are level

We said: The two intervals do not overlap, and one of them finished more of the errands in the set we scored.

We now say: The two rates are close and their intervals overlap almost entirely. On the whole exam they finish errands at about the same rate.

The sentence did not match our own results file or the table printed on the same page.
2026-10-12/pine-vs-town/

Town has no published completion rate

We said: Our scorecard carries a completion rate for both, with an interval and a task count, and the two are not close.

We now say: Our scorecard carries a completion rate for Pine. Town's is withheld, because the figure we were given does not reconcile with the records we can check.

Town's row has been withheld since the scorecard was published. The page's own table showed it as withheld.

Pine is not the only one that phones businesses

We said: Pine phones companies and almost nothing else on this bench does; Pine is the one that calls and Instinct the one that texts; Muse has no answer for errands that end with a company; the channel difference is the whole story for cancellations and disputes.

We now say: Pine is built around calls. TechCrunch reported on 28 September that Instinct has a concierge feature that phones businesses, and Meta's guidance inside Muse says business calling is in early access. We have seen neither of those make a call.

The pages had not caught up with either feature, and the 10-10 Muse calling correction was applied to only part of one page.

The gap between Pine and Instinct is not the widest

We said: The gap between their completion rates is the widest on the board among products with usable figures.

We now say: The two intervals do not come close to overlapping.

Instinct against Muse is a slightly wider gap in our results file.
2026-10-12/agents/town/

Town's Pro trial lasts seven days, not two weeks

We said: The signup trial was described as lasting two weeks.

We now say: The signup Pro trial lasts seven days.

https://www.town.com/pricing, re-read on 12 October 2026.
2026-10-12the scorecard

Pine has the largest published sample: 167 tasks, not 159

We said: The largest scored sample on the board and the widest lead that survives its own interval.

We now say: 159 scored tasks; Pine has the largest published sample at 167. Marginal intervals alone do not establish a significant lead.

Independent round-three audit checked against the cited source and existing teardown evidence; unsupported inference removed.
2026-10-10/muse-vs-pine/

Muse is getting business calling, in early access

We said: Pine phones companies on your behalf and Muse does not.

We now say: Meta's own guidance inside Muse says business calling is in early access and reaching accounts gradually. The account we examined never received it, so we have not seen a Muse call happen.

Guidance that ships inside the product, collected after the comparison was written, describes the calling feature and its limits.

Instinct does have a confirmation step

We said: Living in a message thread costs Instinct the confirmation step the other products rely on.

We now say: A message thread costs it an editable plan. It keeps a confirmation step: an approval card with approve, deny or dismiss, bound to the exact text it showed you.

Our own scorecard already described the approval card. The two pages contradicted it.

Town's template and assistant counts

We said: Town's template runs to 24 sections, and Town has 33 specialised assistants, each with its own instructions.

We now say: The main template we collected has 11 sections, and 24 is only what a fuller version is said to run to. Of 33 assistant records, 29 carry their own instructions and 4 are shells whose work is done in code we never saw.

Our teardown notes were revised after the pages were written. We had printed the claimed figure as if we had seen it.

Other fixes, by page

552 more entries, 485 of them from the review of every published page on 2026-10-12. Most narrow a sentence that went beyond its evidence; the rest correct a detail such as a public benchmark figure or a consumer rule.

Show the 54 pages affected