What changed in AI assistants this quarter

admin
What changed in AI assistants this quarter cover graphic

The past three months have been busy for AI assistants. Anthropic, OpenAI, Google and Microsoft all shipped new models or reworked their main assistant products between August and early November 2026, and several of those changes affect what you pay for, which model you get by default and how much an assistant can do without you.

This article is a dated round-up of what changed, with a source for each item. It is written for people who choose or manage AI tools, so each section ends with what the change means in practice. We have kept to what sources state and have not tried to predict what comes next.

How we compiled this: we did not run hands-on tests. This is based on vendor announcements and public reporting, checked on 11 November 2026. We could not open some OpenAI pages directly, so OpenAI items rely on named secondary sources and are marked as such. Vendor claims, including benchmark figures, are the vendors’ own and have not been independently verified by us.

Key takeaways

  • Anthropic released Claude Opus 5.5 on 22 September 2026, with API prices of $4 per million input tokens and $20 per million output tokens, then Sonnet 5.5 (28 September), Haiku 5.5 (7 October) and Fable 5.2 and Mythos 5.2 (9 November).
  • OpenAI’s GPT-6 reached all ChatGPT users in October with “Intelligent UI”, which builds charts, forms and small tools inside answers; separately it launched always-on agents called dots for top-tier plans.
  • Google said on 24 September that Gemini 4 had entered post-training, and a secondary source reports Gemini 4 Argon reached Google AI Pro and Ultra subscribers on 13 October.
  • Microsoft announced a redesigned Copilot on 9 October with Home, Code and Autopilot, a persistent agent that can keep working without a prompt.
  • The common direction is assistants that act over time, not just answer. That raises practical questions about permissions, usage limits and checking output.

Summary

Three themes stand out across the sources we read.

  • New model generations, quickly. Both Anthropic and OpenAI moved to new numbered generations or point releases within weeks of each other, and Google said its next flagship was close.
  • Assistants that run on their own. OpenAI’s dots and Microsoft’s Autopilot are both described as persistent agents. Google’s Gemini Spark, announced in May, belongs to the same trend, although it is outside our window.
  • Answers that are not just text. ChatGPT’s Intelligent UI renders interactive elements inside the response. Microsoft’s Copilot builds small apps from natural language.

Dates below are those given by the cited source. Where a source did not give a date, we say so.

Anthropic: the Claude 5.5 family

Claude Opus 5.5, 22 September 2026

Anthropic’s announcement says Opus 5.5 was released on 22 September 2026 as the first model in the Claude 5.5 family, and is available on all platforms including Amazon Web Services, Google Cloud and Microsoft Azure. The stated API price is $4 per million input tokens and $20 per million output tokens, which Anthropic says is 20 percent less than Opus 5 ($5 and $25). Anthropic says typical workloads cost about 40 percent less to run than with Opus 5, based on its own tests, and that output is more than 30 percent faster.

For subscribers, the page says five-hour usage limits increased on Pro, Max, Team and seat-based Enterprise plans. Anthropic also states that thinking mode can no longer be switched off, that most cybersecurity tasks are rerouted to Claude Opus 4.8 as a safeguard, and that the model includes watermarking to comply with the EU AI Act. Benchmark figures in the announcement are Anthropic’s own, and the company notes that benchmark margins are a less reliable guide to real-world differences.

Sonnet 5.5, Haiku 5.5 and the 5.2 releases

Anthropic’s newsroom lists “Introducing Claude Sonnet 5.5” on 28 September, “Introducing Claude Haiku 5.5” on 7 October and “Introducing Claude Fable 5.2 and Claude Mythos 5.2” on 9 November. We read the titles and dates on the newsroom page; we did not open the individual pages for these three, so we do not report their details. Microsoft’s October Copilot post says a Sonnet 5.5 model improves slide layouts in PowerPoint, which shows how quickly the model reached third-party products.

Earlier context from Anthropic’s own pages: Claude Fable 5 and Mythos 5 were announced on 9 June 2026, and Opus 5 on 24 July 2026. Both are before the window we cover.

Other Anthropic items

The newsroom also lists, among other items, a 2026 Usage Policy update (8 October) and the donation of Agent Skills to the Agentic AI Foundation (22 October). We have not read those pages and mention them only so you know they exist.

What it means: if you use Claude through the API, the Opus 5.5 price cut is the most concrete change. If you use it through a subscription, check the plan limits yourself, since the usage-limit increase is stated by Anthropic and we have not measured it.

OpenAI: GPT-6, Intelligent UI and dots

Note on sourcing: we could not open OpenAI’s own announcement pages in this research. The items below come from third-party reporting and a third-party timeline of OpenAI’s release notes. Confirm each on OpenAI’s site.

GPT-6 and Intelligent UI, October 2026

VKTR reports that GPT-6 reached paid customers in September and began rolling out to all ChatGPT users on 7 October 2026, in two versions, GPT-6 Sol and GPT-6 Luna. A timeline of OpenAI release notes dates the release of GPT-6 Sol and Luna to 22 September and the Intelligent UI update to 7 October. Intelligent UI lets the model build charts, forms, tappable buttons and small working tools inside a response, choosing the format based on the question. According to Fone Arena, Plus, Pro, Business and Enterprise users get GPT-6 Sol in the Chat tab, Free and Go users get GPT-6 Luna from 8 October, and Enterprise access depends on admin settings. The update applies to the Chat experience only, not Work or Codex.

OpenAI’s own figures, as relayed by VKTR, say GPT-6 Instant starts answering web-search questions 44 percent sooner than GPT-5.6 Instant. That is a vendor evaluation, not independently verified. VKTR also flags that interactive outputs generated on the fly can look authoritative, which raises verification concerns for enterprise teams.

Dots, 29 September 2026

Codersera reports that OpenAI launched dots on 29 September 2026 at DevDay in San Francisco: always-on agents inside ChatGPT, each running on GPT-6 Astra with its own cloud computer and browser, connected to apps through plugins. The article says access is limited to Pro or Business Premium users aged 18 and over, excludes the European Economic Area, Switzerland and the UK on Pro, has no free tier, and that setup is desktop-only. It reports that the first dot is included in those plans and that extra dots had no announced price. Because this comes from one secondary source, treat the details as provisional.

Other ChatGPT changes from the release-notes timeline

Date (2026) Change reported
18 August ChatGPT for Teens launches
17 September ChatGPT becomes available as a Microsoft Word add-in
6 October Audio file uploads for paid users
15 October Chrome tab access and computer use arrive in ChatGPT
20 October Locked chats require a PIN or device authentication
29 October A GPT-6 Chat update makes responses shorter and less flattering for Free, Go and fallback users

A separate search-industry digest also reports that OpenAI closed its ChatGPT Atlas browser in August, less than 10 months after launch, moving its browser proxy function into ChatGPT and Codex. We have not confirmed that on an OpenAI page.

What it means: if your team uses ChatGPT, the default model has changed and answers may now appear as interactive elements. That is useful for comparisons and calculations, but it makes it more important to check AI output for mistakes, especially numbers produced inside a generated chart.

Google: Gemini 4 and the model line-up

Google’s story in this window is mostly about what had not shipped yet. TechCrunch reported that on 21 July Google released Gemini 3.6 Flash, Gemini 3.5 Flash-Lite and a limited-access 3.5 Flash Cyber, without the Gemini 3.5 Pro it had teased in May. On 24 September, 9to5Google reported that Google DeepMind’s Koray Kavukcuoglu said Gemini 4 had entered post-training, that internal testing was under way to power Antigravity, and that he hoped for a release “much earlier” than the end of 2026, with no timeline given.

A later search-news digest lists “Gemini 4 Argon” rolling out globally to Google AI Pro and Ultra subscribers on 13 October, with the Gemini API release notes showing it generally available the same day, and also lists Gemini 4 Argon in AI Mode for those subscribers. We could not open Google’s own release notes, so treat the 13 October date as from a secondary source. The same digest lists Gemini 3.8 Flash and 3.8 Flash Cyber in September. A GitHub changelog entry dated 10 November, “Gemini 4 Carbon in GitHub Copilot”, shows another Gemini 4 variant exists, but we found no further details.

What it means: Google’s model naming is changing fast and the sources we found do not explain how the variants differ. If you rely on Gemini, read the current model documentation before assuming which version a given plan gives you.

Microsoft: Copilot reworked

IT Brief reported that on 9 October 2026 Microsoft announced a redesigned Copilot with Home, Code and Autopilot.

  • Home combines Chat and Cowork as the main entry point, rolling out through the Frontier programme “in the coming weeks”.
  • Code lets people build simple software tools from natural language, using the same underlying technology as GitHub Copilot in a sandboxed environment, with a Copilot Managed Runtime in preview.
  • Autopilot, previously known as Scout, is a persistent agent that can keep working without a prompt. Users assign it a name, role and objective, and it includes permissions, audit and governance controls. IT Brief reports it moving to private preview by the end of the month.

Microsoft’s own “What’s new in Microsoft Copilot, October 2026” post adds detail on rollout stages: some features reached Frontier users in October with worldwide rollout in November, and others, such as Office files in Workspaces, are listed for January. It also lists Projects, skills in Copilot Chat, role-based plugins for Sales, Service, Education, Legal and Finance, a legal plugin with more than 20 skills in Word, and a Purview data loss prevention policy covering Copilot Chat and Cowork in public preview.

What it means: Microsoft staggers rollouts by programme and by month, so a feature announced in October may not be in your tenant yet. Admins should check the specific rollout stage before promising a capability to staff, and read our guidance on protecting company data before enabling agents that act on mailboxes and files.

Other assistant news

A search-industry digest reports that on 20 October Perplexity added a local-first mode that lets its Computer agent run on a Mac, including offline, for Pro, Max and Enterprise subscribers with at least 36 GB of unified memory, citing press reports. On 5 October it reports automations and inline charts in Perplexity Computer, citing third-party trackers. These are second-hand reports and we did not confirm them on Perplexity’s site.

The same digest notes ChatGPT Ads in 89 countries in October. That is a business-model change worth watching for anyone who uses free tiers for work.

Timeline at a glance

Date (2026) Vendor Change Source type
22 September Anthropic Claude Opus 5.5 released Vendor announcement
22 September OpenAI GPT-6 Sol and Luna released Third-party timeline
24 September Google Gemini 4 said to be in post-training 9to5Google report
28 September Anthropic Claude Sonnet 5.5 listed Vendor newsroom
29 September OpenAI Dots launched at DevDay Secondary report
7 October OpenAI GPT-6 with Intelligent UI rolls out in ChatGPT Secondary reports
7 October Anthropic Claude Haiku 5.5 listed Vendor newsroom
9 October Microsoft Copilot redesign: Home, Code, Autopilot IT Brief report
13 October Google Gemini 4 Argon reported for AI Pro and Ultra Search digest
9 November Anthropic Claude Fable 5.2 and Mythos 5.2 listed Vendor newsroom

What it means for tool buyers

  • Re-check defaults. Default models changed for ChatGPT’s free and paid tiers, and Anthropic made thinking mode non-optional on Opus 5.5. If you compared assistants earlier in the year, that comparison may be out of date. Our guide on comparing AI tools fairly explains how to redo it.
  • Look at limits as well as prices. Several announcements talk about usage limits, credits and gradual rollouts. See how to read an AI tool’s pricing page.
  • Treat agents as a permissions decision. Persistent agents with their own accounts, browsers or mailboxes need clear rules about what they may touch.
  • Expect errors to persist. Newer models do not remove the need to verify. See what hallucinations are and how to reduce them.

What to watch next

These are open questions drawn from the sources, not predictions.

  • Whether Google publishes details on how its Gemini 4 variants differ and which plans get which.
  • How OpenAI prices additional dots, since the secondary source we read says no price was announced.
  • When Microsoft’s Autopilot and Code move from Frontier and preview programmes to general availability.
  • What the Fable 5.2 and Mythos 5.2 releases contain, which we did not read.

Frequently asked questions

What is the newest Claude model?

Anthropic’s newsroom lists Claude Fable 5.2 and Claude Mythos 5.2 on 9 November 2026. We did not read the details. Claude Opus 5.5, released on 22 September, is the one we confirmed on a vendor page.

Did ChatGPT switch to GPT-6 for free users?

Fone Arena reports that Free and Go users get GPT-6 Luna from 8 October 2026. Confirm in ChatGPT’s own release notes.

What is Intelligent UI?

A ChatGPT capability described by VKTR as letting the model build charts, forms, buttons and small tools inside a response.

Is Gemini 4 released?

A secondary source reports Gemini 4 Argon for Google AI Pro and Ultra subscribers from 13 October 2026. On 24 September a Google DeepMind executive said Gemini 4 was in post-training. We could not confirm the details on Google’s own pages.

What is Microsoft Autopilot?

According to IT Brief, a persistent cloud-hosted agent, previously called Scout, that can keep working on an objective without a prompt, with governance controls.

Which assistant is best?

We cannot say. We have not tested these products, and the answer depends on your tasks, plan and data rules. Use our guides to run your own comparison.

Conclusion

Between August and early November 2026 the main assistants moved toward newer model generations and toward agents that work in the background. The details are still settling, and several important announcements reach users in stages. Before changing tools, read how to choose an AI tool, check how to protect company data, and test on your own work using our trial checklist.

Sources

Keep reading