CHOOSELY
Monthly AI-tool changes report
Version 1.0
29 Verified AI-Tool Changes That Mattered in October 2026
What changed in October, what you can use now, and what an existing subscription buys.
- Period
- October 1 to October 31, 2026
- Verification
- Verified through November 1, 2026, 01:31 UTC (12:31 AEDT)
- Author
- Choosely Editorial
- Evidence
- Evidence-based analysis
The HTML report is the canonical record. Material corrections are dated and reflected across HTML, PDF and CSV.
Key findings
29
material changes verified
10
launches and announcements
8
capability changes
1
effective retirement
October made AI feel more like a working environment. Office editing moved inside Copilot, ChatGPT added collaborative slides, and coding tools gained new ways to coordinate projects and operate applications. The useful comparison is increasingly the finished document or completed task.
Cheaper models widened the set of jobs worth trying. Haiku 5.5 and GPT-6.1 Terra joined the options for bounded work, while Sonnet 5.5 made it cheaper to reuse previously stored input, known as cache reads. A cheaper token is useful only if the task still gets done well.
Subscriptions became harder to compare by price alone. Credits for developers calling models from their own apps, Work/Codex allowance increases, Notion’s split balances and Claude Ultra’s verified-account session rules all changed what an existing payment could buy. These are different kinds of capacity, with different limits.
The most impressive announcement was not always the product a reader could use. Carbon, Genie 4, Duet and Rosalind each had restricted access or preview conditions. The report counts verified changes without turning a waitlist into general availability.
October at a glance
| Primary change class | Changes | Share |
|---|---|---|
| Model or product launch | 10 | 34% |
| Availability or access change | 1 | 3% |
| Capability change | 8 | 28% |
| Plan change | 6 | 21% |
| Price change | 1 | 3% |
| Policy change | 2 | 7% |
| Deprecation | 0 | 0% |
| Shutdown | 1 | 3% |
One primary class per event. A launch can have pricing or retirement consequences without being counted twice.
Coverage statement
This report selects material changes released, announced or effective during October 1–31, 2026. Verification closed November 1 at 01:31 UTC (12:31 AEDT). Dates follow the vendor’s dated announcement unless an event explicitly records an observed-by date or a later effective phase.
The 29 records span ten vendor ecosystems: OpenAI, Anthropic, Google, GitHub, Microsoft, Notion, Canva, SpaceXSI, Cursor and Perplexity. Selection is based on a concrete change to access, cost, workflow, control or migration; it is not a census of the market.
Copilot’s global policy was announced in September. October’s record captures its documented enforcement phase and links SEP-2026-014. Count it once within October, but deduplicate the shared transition before combining monthly datasets. Announced restricted launches and rollout completion are distinguished throughout.
The 29 verified changes
Grok Bot · Browser access
Grok Bot reached the browser for more subscribers
The October browser rollout made ongoing Bot conversations accessible on grok.com. Subscribers could continue their Bot conversations from a browser instead of relying on another client.
- Previous state
- Bot access existed on other supported surfaces.
- New state
- SuperGrok Plus and Heavy subscribers could use Bots on grok.com.
- Who is affected
- Eligible Grok subscribers.
- Decision effect
- Check browser access before changing plans.
Announced or released . Effective: October 3, 2026. Verified .
Vendor-dated release or rollout start; not proof that every account had access.
Perplexity Computer · Automations
Perplexity Computer gained ongoing event-driven work
October’s release documents scheduled and event-triggered work using memory, skills and connected apps. It expands the ongoing-work proposition; it is not the first evidence of recurring Computer tasks.
- Previous state
- Scheduled automations were documented in August; the exact earlier event-trigger scope is not established.
- New state
- Automations could run on schedules or connected-app events on the web and pause for consequential-action review.
- Who is affected
- Eligible Computer users with supported connections.
- Decision effect
- Start with a bounded recurring task and inspect its review behavior.
Announced or released . Effective: October 5, 2026. Verified .
Vendor-dated release or rollout start; not proof that every account had access.
ChatGPT · Audio-file uploads
ChatGPT added audio-file conversations
Recordings could become source material for a conversation. This creates a useful route from audio to notes, while transcription remains fallible.
- Previous state
- The exact earlier audio-file support state was not reconstructed; October 6 is the dated announcement of this workflow.
- New state
- Paid subscriptions and workspaces gained audio uploads for transcription and follow-up questions, subject to eligibility.
- Who is affected
- People working with interviews, meetings and lectures.
- Decision effect
- Check important transcript passages against the recording.
Announced or released . Effective: October 6, 2026. Verified .
Vendor-dated release or rollout start; not proof that every account had access.
ChatGPT · GPT-6 and Intelligent UI in Chat
GPT-6 brought interactive answers into ordinary Chat
This was a new everyday work surface for existing models, not a second Sol or Luna model launch. A question could produce a usable comparison or small interactive tool instead of only prose.
- Previous state
- GPT-6 Sol and Luna had already launched; this Chat experience had not.
- New state
- Chat began combining answers with interactive visuals; paid rollout started October 7 and Free/Go expansion October 8. Pro reasoning was excluded.
- Who is affected
- Chat users in supported plans and workspaces.
- Decision effect
- Evaluate the new Chat experience separately from Work and Codex.
Announced or released . Effective: October 7, 2026. Verified .
Vendor-dated release or rollout start; not proof that every account had access.
Claude Haiku 5.5 · Small model family
Haiku 5.5 made small-model routing worth revisiting
Summaries, classification and supporting agent work gained a new lower-cost candidate. Try it on high-volume work whose success is easy to check before moving more demanding tasks.
- Previous state
- Haiku 4.5 was the preceding Haiku generation.
- New state
- Haiku 5.5 launched with adjustable effort and lower token rates; prices differ above 100,000 prompt tokens.
- Who is affected
- Developers routing frequent, bounded requests.
- Decision effect
- Test quality and prompt-size pricing before moving traffic.
Announced or released . Effective: October 7, 2026. Verified .
Vendor-dated release or rollout start; not proof that every account had access.
Pricing basis: USD · per million tokens
Claude Sonnet 5.5 · API cache reads
Sonnet 5.5 cache reads became half the price
Repeated context can now cost less to read. Cache reads reuse previously stored input; the saving does not apply to every part of a request.
- Previous state
- Cache reads cost US$0.20 per million tokens.
- New state
- Cache-read rates fell to US$0.10 per million tokens.
- Who is affected
- Developers repeatedly reading cached context.
- Decision effect
- Estimate savings from actual cached traffic.
Announced or released . Effective: October 7, 2026. Verified .
Vendor-dated release or rollout start; not proof that every account had access.
Pricing basis: USD · per million cached input tokens
Claude · Subscriber API-credit benefit
Claude subscriptions gained a separate API budget
The October 7 release note dates the benefit. Documentation now uses the later Ultra name. Grants fund the Claude Platform, expire without rollover and require activation.
- Previous state
- Eligible subscriptions did not include this new monthly API grant.
- New state
- Max and Team API credits began rolling out; current terms list US$100 Max, US$200 Ultra and up to US$500 pooled Team credits.
- Who is affected
- Eligible subscribers building API apps.
- Decision effect
- Link the intended Console organization; these credits do not top up Claude Code.
Announced or released . Effective: October 7, 2026. Verified .
Vendor-dated release or rollout start; not proof that every account had access.
Pricing basis: USD · monthly API credits
Microsoft Copilot · Office in the Copilot app
Office work moved inside Copilot
Documents, spreadsheets and slides became a more continuous workflow in the Copilot app. The announcement is a Frontier rollout, not universal availability.
- Previous state
- Users moved from Copilot drafting to separate Office editing surfaces.
- New state
- Microsoft began an early-access Frontier rollout of Word, Excel and PowerPoint creation, editing and collaboration inside Copilot.
- Who is affected
- Eligible Frontier customers.
- Decision effect
- Confirm tenant access and review exported files before adopting the workflow.
Announced or released . Effective: October 9, 2026. Verified .
Vendor-dated release or rollout start; not proof that every account had access.
Gemini 4 Argon · Gemini app and developer API
Gemini 4 Argon became a model people could use
Argon entered supported consumer and developer surfaces. Google positions it for difficult work, but a vendor launch does not establish reliability on every task.
- Previous state
- The Gemini family lacked the newly released Argon model.
- New state
- Argon launched with Pro/Ultra app access and developer API access at introductory US$2 input and US$10 output per million tokens.
- Who is affected
- Eligible subscribers and API developers.
- Decision effect
- Separate app entitlements from API billing and introductory rates.
Announced or released . Effective: October 13, 2026. Verified .
Vendor-dated release or rollout start; not proof that every account had access.
Pricing basis: USD · introductory per million tokens
Gemini · Antigravity in the macOS app
Gemini on Mac gained an agentic coding handoff
The integration connects an idea in chat to an agent building and testing code in the background. It is a distinct app workflow, not another Argon launch.
- Previous state
- The Mac chat app lacked this built-in Antigravity handoff.
- New state
- Eligible adults could start an Antigravity coding agent from Gemini on Mac, approve its plan and return to chat while it worked.
- Who is affected
- English-language users in supported Antigravity regions.
- Decision effect
- Review generated code and tests before release.
Announced or released . Effective: October 13, 2026. Verified .
Vendor-dated release or rollout start; not proof that every account had access.
Grok Voice 3.0 · Speech-to-speech API
Grok Voice 3.0 changed the model behind live calls
The API launch kept the preceding minute rate while changing the model. Claims about latency and call success are vendor claims, not Choosely call tests.
- Previous state
- Grok Voice 2.0 powered the prior latest alias.
- New state
- Voice 3.0 launched at the same US$0.08 per minute; grok-voice-latest moved automatically.
- Who is affected
- Voice-agent developers and operators.
- Decision effect
- Retest live call handling when using an automatically updated alias.
Announced or released . Effective: October 14, 2026. Verified .
Vendor-dated release or rollout start; not proof that every account had access.
Pricing basis: USD · per minute
GPT-5.5 · ChatGPT, Work and Codex
GPT-5.5 left ChatGPT while its API remained
A model retirement is product-specific. This record does not imply an API shutdown or retirement of specialized Rosalind variants.
- Previous state
- GPT-5.5 remained available in these ChatGPT surfaces.
- New state
- OpenAI retired GPT-5.5 from ChatGPT, Work and Codex; API access was unaffected.
- Who is affected
- Users relying on the older model in ChatGPT.
- Decision effect
- Retest saved workflows on a supported model.
Announced or released . Effective: October 14, 2026. Verified .
Vendor-dated release or rollout start; not proof that every account had access.
ChatGPT Slides · Presentations in Space
ChatGPT added collaborative presentation work
A deck could stay editable with teammates in Space and export from the web. Eligibility remained narrower than every ChatGPT plan.
- Previous state
- ChatGPT lacked this newly announced collaborative Slides surface.
- New state
- Slides began rolling out to eligible Pro, Business and Enterprise users with Space access.
- Who is affected
- Teams making presentations in supported accounts.
- Decision effect
- Check export fidelity and access before moving a team workflow.
Announced or released . Effective: October 16, 2026. Verified .
Vendor-dated release or rollout start; not proof that every account had access.
Notion AI · Standard and Select credits
Notion split included AI usage into two allowances
The new system was live for new customers in October. The existing-customer migration belongs to November; it has not been counted here as already completed.
- Previous state
- The prior funding system did not provide these two new seat allowances.
- New state
- New Business/Enterprise customers received 1,500 Standard and 500 Select monthly credits per member; existing customers were scheduled for November 2.
- Who is affected
- Notion Business/Enterprise members and admins.
- Decision effect
- Separate included allowances from purchased workspace credits.
Announced or released . Effective: October 19, 2026. Verified .
October 19: new customers live. Existing customers scheduled November 2; not an October effective migration.
Pricing basis: credits · monthly per eligible seat
Cursor Agent SDK · Agents as deployable code
Cursor made reusable agents a deployable software project
The SDK packages instructions, tools, skills, connections and channels as a project with evaluations. Availability is verified; Choosely has not deployed an SDK agent for this report.
- Previous state
- Cursor did not offer this generally available Agent SDK release.
- New state
- The SDK became generally available for defining, evaluating and deploying agents to Cursor or a team’s own infrastructure.
- Who is affected
- Developers building reusable agents.
- Decision effect
- Evaluate tools and event handling before deployment.
Announced or released . Effective: October 20, 2026. Verified .
Vendor-dated release or rollout start; not proof that every account had access.
Claude Duet · Proactive agent research preview
Claude Duet opened a waitlisted proactive-agent preview
Duet can keep working on a cloud computer and return with updates. Its appeal is follow-through across ongoing work; waitlist, permissions and shared usage still govern access.
- Previous state
- Claude did not offer the new Duet app preview.
- New state
- A waitlisted research preview opened, starting with US Pro, Max and Ultra subscribers and sharing Claude usage limits.
- Who is affected
- Eligible professionals and small-business users.
- Decision effect
- Treat it as limited preview access, not a universal plan entitlement.
Announced or released . Effective: October 21, 2026. Verified .
Vendor-dated release or rollout start; not proof that every account had access.
GitHub Copilot · Global default for eligible features
Copilot enforced the policy announced in September
This is the effective phase of SEP-2026-014, not a new announcement. GitHub documented a staged rollout and a one-time policy application, not a recurring rewrite of explicit choices.
- Previous state
- Admins could configure the global policy, but the announced enforcement had not started.
- New state
- GitHub began enforcement October 22 and scheduled rollout through October 26, applying the default to eligible Unconfigured settings while preserving explicit choices.
- Who is affected
- Copilot Business/Enterprise administrators.
- Decision effect
- Audit resolved settings, especially MCP and feature access.
Announced or released . Effective: October 22, 2026. Verified .
Announced/configurable September 24. October enforcement began October 22 and was scheduled through October 26. Per-account completion was not verified. Deduplicate this transition in cross-month totals.
Canva Craft · Design-focused image model
Canva Craft brought a design-first image model into Canva
The useful promise is a better starting composition for a real design. Canva describes design-specific training and editable output; Choosely has not tested the model.
- Previous state
- Canva used an existing mix of in-house and partner image models.
- New state
- Canva introduced Craft in Canva and its public API; the Connect image-generation endpoint remained a preview API.
- Who is affected
- Creators already assembling designs in Canva.
- Decision effect
- Compare the finished design and editing effort, not only the raw image.
Announced or released . Effective: October 22, 2026. Verified .
Vendor-dated release or rollout start; not proof that every account had access.
Claude · Auto approval defaults across apps
Claude announced wider use of automatic action approvals
The dated October update expands the approval-policy story beyond Code. No exact completion date or universal account migration is established here; an automated action check is not a promise that every action asks a person.
- Previous state
- The earlier default change concerned Claude Code.
- New state
- Anthropic announced Auto would roll out the following week as the default for new sessions in most Claude apps on Pro, Max and Team.
- Who is affected
- Users delegating actions in eligible Claude apps.
- Decision effect
- Check the actual conversation mode and connector permissions.
Announced or released . Effective: Not fixed / see date note. Verified .
October 22 announcement says following week. Exact rollout completion and account-by-account state unverified.
GPT-6.1 Terra · API, ChatGPT Work and Codex
GPT-6.1 Terra made routine computer work cheaper to attempt
A lower-cost model can be valuable when the job has a clear finish line. Its launch price is counted within this model event; lower token prices alone do not guarantee lower total task cost.
- Previous state
- GPT-5.6 Terra preceded this model.
- New state
- Terra launched at US$1 input and US$5 output per million tokens; Work/Codex availability is separate from ordinary Chat.
- Who is affected
- Developers and eligible Work/Codex users.
- Decision effect
- Try bounded writing, visual and computer tasks before rerouting complex work.
Announced or released . Effective: October 26, 2026. Verified .
Vendor-dated release or rollout start; not proof that every account had access.
Pricing basis: USD · per million tokens
Genie 4 · Interactive world model
Genie 4 offered a glimpse beyond the generated video clip
The new world model can generate an environment as a person moves through it. This is a restricted announcement, not proof that ordinary Ultra subscribers could already use Genie 4.
- Previous state
- Project Genie used the preceding world-model experience.
- New state
- Google announced Genie 4 with audio and real-time steering for trusted testers; wider Project Genie and API access was planned.
- Who is affected
- Creative and research users assessing interactive environments.
- Decision effect
- Do not buy a subscription on an undated promise of future Genie 4 access.
Announced or released . Effective: Not fixed / see date note. Verified .
Announced October 27; trusted testing confirmed, broader release date unannounced.
Gemini 4 Carbon · Restricted frontier model
Gemini 4 Carbon arrived as a restricted frontier announcement
Carbon points toward longer engineering work, but the launch was not a consumer subscription upgrade. Google’s capability claims remain attributed rather than treated as Choosely test results.
- Previous state
- Argon was the newly available public Gemini 4 model.
- New state
- Google announced Carbon with a staged rollout to selected cyber defenders and Fairwind partners; broad access remained undated.
- Who is affected
- Vetted defenders and people comparing frontier claims.
- Decision effect
- Keep public Argon access separate from restricted Carbon.
Announced or released . Effective: Not fixed / see date note. Verified .
Announced October 28; staged trusted rollout that week, no general-access date.
ChatGPT · Subscriber API-credit benefit
ChatGPT subscriptions gained a separate API-credit benefit
The October 28 release note dates the offer. Activation, billing and request limits remain separate from the ordinary ChatGPT allowance.
- Previous state
- Eligible subscriptions did not include this new promotional API benefit.
- New state
- Eligible Pro and Business subscribers could activate API credits for a linked organization, with plan, subscription-tenure and organization-ownership conditions.
- Who is affected
- Subscribers building API apps.
- Decision effect
- Verify the grant, expiry and covered API products; it is not extra signed-in Codex usage.
Announced or released . Effective: October 28, 2026. Verified .
Vendor-dated release or rollout start; not proof that every account had access.
Pricing basis: USD · subscription-linked API credits
GitHub Copilot Fusion · Model routing and orchestration
Copilot Fusion made model routing part of the product
Fusion chooses and coordinates models for work. Instead of choosing a model for every step, developers can let the product route the work while retaining effort and administrator controls.
- Previous state
- Users did not have the newly released Fusion routing system.
- New state
- Fusion became generally available across supported Copilot clients and all Copilot plans, with administrator model policies still applying.
- Who is affected
- Developers choosing how Copilot routes work.
- Decision effect
- Inspect model cost and test representative tasks before adopting a routing preset.
Announced or released . Effective: October 28, 2026. Verified .
Vendor-dated release or rollout start; not proof that every account had access.
GitHub Copilot app · Projects and desktop computer use
Copilot’s app added project and desktop-action previews
The grouped app update changes where coding work can reach. It is a public preview rather than a blanket authorization to operate a computer.
- Previous state
- The app lacked these newly announced project and computer-use previews.
- New state
- GitHub opened Projects and desktop computer use in public preview; computer use is opt-in with app approval and admin controls.
- Who is affected
- Copilot app users on supported macOS and Windows systems.
- Decision effect
- Limit approved applications and review actions on the desktop.
Announced or released . Effective: October 28, 2026. Verified .
Vendor-dated release or rollout start; not proof that every account had access.
GitHub Enterprise · Bundled Copilot access
GitHub Enterprise bundled access, not an AI-usage allowance
Removing a separate access purchase changes procurement. It does not create a free usage pool or establish the same offer for self-serve customers.
- Previous state
- These Enterprise customers needed a separate Copilot access purchase.
- New state
- Account-team GitHub Enterprise Cloud customers gained Copilot access, but no included AI credits; usage is metered from the first request.
- Who is affected
- Enterprise Cloud customers working with GitHub or Microsoft account teams.
- Decision effect
- Compare metered spend with existing seat arrangements; self-serve access is not established.
Announced or released . Effective: October 28, 2026. Verified .
Vendor-dated release or rollout start; not proof that every account had access.
ChatGPT Pro · Work and Codex allowance
ChatGPT Pro gained more Work and Codex capacity
The change alters how much work the subscription funds. It is separate from API credits and does not imply unlimited usage.
- Previous state
- Pro tiers had lower or differing Work/Codex allowances and Pro 200 faced a planned reduction.
- New state
- Pro 100/500 allowances doubled; the Pro 200 reduction was canceled and lower-allowance Pro 200 subscribers were raised.
- Who is affected
- Heavy individual Work and Codex users.
- Decision effect
- Compare the allowance now included before upgrading.
Announced or released . Effective: October 29, 2026. Verified .
Vendor-dated release or rollout start; not proof that every account had access.
GPT-6 Rosalind Research · Restricted life-sciences model
GPT-6 Rosalind Research entered a limited scientific rollout
This specialized model is not available in Chat. The prior Research model remained available, and Discovery stayed on 5.5.
- Previous state
- Approved Research users had GPT-5.5 Rosalind Research.
- New state
- GPT-6 Rosalind Research began a limited rollout in Work, Codex and the API; specific Research approval is required.
- Who is affected
- Approved life-sciences organizations and authorized researchers.
- Decision effect
- Confirm Research authorization; Discovery approval is insufficient.
Announced or released . Effective: October 30, 2026. Verified .
Vendor-dated release or rollout start; not proof that every account had access.
Claude Max and Ultra · Individual subscription names and limits
Claude’s $200 tier became Ultra with a different session limit
The new terms were recorded by Choosely on October 31 Sydney time and freshly rechecked. Weekly and model-specific limits remain. The exact global announcement time is not independently established here.
- Previous state
- The two premium plans were named Max 5x and Max 20x.
- New state
- The names became Max and Ultra at unchanged US web monthly prices of US$100/200; verified Ultra accounts have no five-hour limit.
- Who is affected
- Existing and prospective heavy Claude users.
- Decision effect
- Check verification and weekly limits; Ultra is not unlimited.
Observed by . Effective: Not fixed / see date note. Verified .
October 31 is the retained Sydney observation date, not a claimed universal release instant. Exact account rollout time unknown.
Pricing basis: USD · US web per month
Primary source register
- OCT-2026-001 · Official announcement
- OCT-2026-002 · Official announcement · Prior scheduled-automation state (August 24)
- OCT-2026-003 · Official announcement
- OCT-2026-004 · Official announcement
- OCT-2026-005 · Official announcement
- OCT-2026-006 · Official announcement
- OCT-2026-007 · Official announcement · Dated Claude release notes
- OCT-2026-008 · Official announcement
- OCT-2026-009 · Official announcement
- OCT-2026-010 · Official announcement
- OCT-2026-011 · Official announcement
- OCT-2026-012 · Official announcement
- OCT-2026-013 · Official announcement
- OCT-2026-014 · Official announcement
- OCT-2026-015 · Official announcement
- OCT-2026-016 · Official announcement
- OCT-2026-017 · Official announcement
- OCT-2026-018 · Official announcement · Dated Canva developer announcement
- OCT-2026-019 · Official announcement
- OCT-2026-020 · Official announcement · Dated model release notes
- OCT-2026-021 · Official announcement
- OCT-2026-022 · Official announcement
- OCT-2026-023 · Official announcement · Dated ChatGPT release notes
- OCT-2026-024 · Official announcement
- OCT-2026-025 · Official announcement
- OCT-2026-026 · Official announcement
- OCT-2026-027 · Official announcement
- OCT-2026-028 · Official announcement
- OCT-2026-029 · Observed Ultra plan terms · Max plan terms
Pricing and plan analysis
- 1
Keep the balances separate
Subscription allowance, API credits and purchased workspace credits are not interchangeable. Check where each balance can be spent and when it expires before paying for more.
- 2
Read the unit beside the saving
Sonnet’s price change concerns cache reads. Terra’s launch rates concern API tokens. Voice 3.0 retains a minute rate. Compare the cost of the workload you actually run.
- 3
Included access can still mean metered usage
GitHub’s Enterprise access bundle does not include an AI-credit allowance. Claude Ultra’s removal of a session window for verified accounts still leaves weekly limits. The headline alone cannot tell you the next bill or lockout.
Category analysis
1. The work surface became part of the product
A familiar example is a client presentation. One system can draft the argument, another can edit the deck alongside colleagues, and a coding agent can build a supporting dashboard. October’s office and app updates bring more of that work into the same conversation or project. That can remove handoffs even when the underlying model is familiar.
The test is what survives outside the demo: an editable file, a working application, a clear source trail and a result another person can inspect. No hands-on comparison is claimed here. The suggested evaluation is to give a shortlisted product one representative job and check the artifact in its destination application.
2. Routing mattered as much as a flagship name
Haiku and Terra strengthen the case for sending straightforward work to a lower-cost model. Fusion makes choosing and coordinating models, known as routing, a product feature, while Notion’s funding structure gives automatic and manually selected work different balances. Buyers now face a system decision as well as a model decision.
A sensible trial measures corrections, retries and human review time alongside the API bill. A low rate can lose its advantage if a difficult job has to be repeated. Conversely, a flagship can be unnecessary for a well-defined extraction or first draft.
3. Persistent agents changed the approval question
Duet, Perplexity Automations and the new app workflows promise less repeated briefing. The agent can remember the assignment, wait for an event or continue while its owner is elsewhere. That makes delegated follow-through the attraction, rather than the novelty of a chatbot that can click.
It also makes the boundary of the assignment more consequential. Connecting an account, choosing an approval mode and permitting spending are separate decisions. Start with a bounded job, inspect the first results and expand access only when the workflow earns it. A vendor’s safety evaluation is not a guarantee for every connected account.
What to watch in November
Future checks, not October event totals
- Notion’s November 2 transition for existing customers: reconcile the actual allowances and funding route after migration.
- Carbon and Genie 4: look for a dated broader release and real account eligibility before treating announced access as available.
- Claude Auto and Ultra: check account-level rollout, verification and remaining limits without inferring universal completion.
- Arc migration deadlines and Custom GPT retirement: check current account notices before relying on a future date. These are watchlist checks, not October shutdown counts.
Methodology
What earns a record
A change must affect access, workflow, cost, control or migration. We surveyed dated official sources across the month, beyond Choosely’s articles. Minor fixes, promotional examples and rumors do not qualify.
Each event has one primary class. Related model variants stay together; a launch’s own pricing does not add a count. A separate plan benefit, existing-model price cut or distinct workflow can qualify independently.
Dates and evidence
OCT-2026-NNN records retain announcement/release, effective and verification dates, previous/new states, affected users, decision effect and primary links. A blank effective date means no exact date is established. Date notes qualify phased access and Ultra’s retained October 31 Sydney observation.
Copilot enforcement follows SEP-2026-014. Its shared transition key must be deduplicated before adding monthly totals. Vendor announcements establish announced terms, not every account’s access or the outcome of an ordinary task.
One dataset, three formats
One typed dataset drives HTML cards, totals, CSV and PDF, preserving IDs and cutoff. HTML is canonical. Performance claims remain attributed; no controlled Choosely product tests are claimed.
Version 1.0 published November 1, 2026. The verification cutoff remains November 1 at 01:31 UTC; publication does not imply a later source review.
Limitations
- Selection across ten ecosystems is not a market census or evidence that AI accelerated relative to September.
- We did not test account entitlements, billing or product performance. Restricted launches and rollout completion remain qualified.
- Help pages can change in place. Dated sources and retained observations support chronology, with unknowns disclosed in date notes.
- Terms can change after the cutoff; region, tax, app-store and negotiated pricing may differ. Check current terms before buying or migrating.
Downloads and reuse terms
The CSV contains the 29 event-level records used here, including source URLs, dates, prior and new state, affected users and decision effect. The PDF presents the same records for reading and archiving.
© 2026 Choosely. All rights reserved. The dataset may be quoted and referenced with attribution. Redistribution, republication or commercial reuse of the complete dataset requires written permission from Choosely.
Corrections and version history
Material corrections will be dated and explained across all three formats. Stable event IDs remain attached to corrected or superseded records. Factual correction requests can be sent to info@choosely.ai.
Version 1.0 published November 1, 2026. Initial publication; the evidence cutoff is unchanged.
Final October verdict
October’s best upgrade may already be inside the tool you pay for. Test the new work surface, cheaper model or allowance on one recurring job before buying another subscription. Judge the finished work, keep spending and approval boundaries clear, and treat restricted frontier announcements as something to follow rather than something you already own.
Choosely Editorial. 29 Verified AI-Tool Changes That Mattered in October 2026. Choosely, November 1, 2026. Version 1.0. https://choosely.ai/reports/ai-tool-changes/2026-10