What changed in AI coding tools this quarter

admin
What changed in AI coding tools this quarter cover graphic

AI coding tools changed in three ways between August and early November 2026: ownership and supply deals shifted around one of the best-known editors, the big tools added more autonomous cloud agents and app-building features, and the biggest model vendors released new models that the coding tools adopted within days.

This article is a dated round-up with a source for each item, written for developers and for managers who pay for these tools. We stick to what sources state, mark vendor claims as claims, and say where we could not verify something.

How we compiled this: we did not run hands-on tests. This is based on vendor changelogs, announcements and public reporting, checked on 11 November 2026. Several release digests did not give dates for individual items, and we say so where that applies.

Key takeaways

  • SpaceX completed its acquisition of Cursor’s parent company, Anysphere, on 14 August 2026, according to a report, and OpenAI proposed ending its model supply to Cursor from 12 November 2026.
  • Cursor’s changelog lists an Agent SDK (20 October), live app building and hosting (5 November) and voice chat with agents (6 November).
  • GitHub’s changelog shows Copilot arriving in JetBrains Air (3 November) and Xcode’s coding assistant (4 November), and new models added within a day or two of release.
  • Anthropic released Claude Opus 5.5 on 22 September and says it leads on agentic coding benchmarks, a claim that is Anthropic’s own.
  • Agents now run in the cloud with their own environments, which makes permissions, secrets and review practices more important.

Summary

The sources we read describe a market where model choice, agent autonomy and enterprise controls all moved at once.

  • Models: new Claude and Gemini models appeared in GitHub Copilot on the day or the day after release, according to GitHub’s changelog titles.
  • Agents: Cursor, Devin and OpenAI’s Codex CLI all shipped changes aimed at running several agents or long-lived sessions.
  • Controls: Devin added SSO and budget hierarchies, and Cursor says live apps must pass checks for exposed secrets before going live.

Cursor: acquisition, supply and new features

Ownership and OpenAI supply

You.com’s news feed reports that SpaceX completed its acquisition of Anysphere, Cursor’s parent, on 14 August 2026, in an all-stock deal with an implied equity value of about $60 billion. It says OpenAI gave notice that it would stop supplying its models to Cursor, proposing 12 November 2026 as the cutoff, subject to negotiation, and that OpenAI said it would not provide future models during the wind-down. The article quotes OpenAI as saying it “cannot be confident” SpaceX will use its technology within its terms of service. A timeline of OpenAI release notes dates the announcement to 28 August.

The same article says Cursor is not shutting down, that Cursor’s CEO said OpenAI models account for about 5 percent of user traffic (a figure with no disclosed method), and that Cursor offers Grok 4.5 and 4.6 developed with SpaceXAI, its own Composer models, Anthropic’s Claude and Google’s Gemini. It also reports that Anthropic co-founder Tom Brown said Anthropic will keep increasing compute for Claude models in Cursor. Coverage framed the dispute partly as part of a wider feud between the companies’ leaders, while some outlets called it a business decision. We report both without taking a side.

Cursor’s changelog, October and November

Date (2026) Cursor changelog entry
13 October Origin App Marketplace
16 October Cloud Agent Canvases and Environment Improvements
20 October Agent SDK
5 November Build Live Apps in Cursor
6 November Voice Chat with Agents

The changelog page we opened showed no entries for August or September, and only titles for these five. A release digest for 2 to 8 November describes two of them in more detail. Realtime voice chat lets agents take voice calls in the Agents Window, the IDE, the web and iOS, runs on Grok Voice with selectable voices, and has zero data retention under Privacy Mode. Live app building lets cloud agents scaffold a web app, provision Postgres, add Cursor SSO authentication and deploy to a shareable link on a custom domain, and the digest says checks for exposed secrets must pass before anything goes live.

What it means: if your team uses OpenAI models inside Cursor, ask which models you will have after 12 November. If you are evaluating Cursor, note that the model line-up depends on supply deals that can change.

GitHub Copilot

GitHub’s changelog page lists a run of Copilot entries in early November 2026:

Date (2026) Entry
3 November GitHub Copilot is now available in JetBrains Air
3 November BYOK and enterprise custom models in the Copilot usage API
4 November GitHub Copilot is now available in Xcode’s coding assistant
6 November Copilot impact dashboard adds chat and VS Code agent usage
6 November @GitHub in Slack and Teams is generally available
9 November Claude Fable 5.2 in GitHub Copilot
10 November Gemini 4 Carbon in GitHub Copilot

We read titles and dates on the changelog index, not each entry, so we cannot say which plans get each model or what limits apply. We also saw no Copilot entries listed for October on that page, which may reflect how the page displays older months.

What it means: GitHub is positioning Copilot as a multi-model, multi-editor layer, and adding usage reporting for managers. Check model availability and any premium request limits on GitHub’s documentation before you rely on a specific model.

Anthropic and OpenAI

Anthropic

Anthropic’s Claude Opus 5.5 announcement, dated 22 September 2026, says the model leads on agentic coding, computer use and knowledge-work benchmarks, citing 66.4 percent on Terminal-Bench 4.0 against 52.3 percent for Opus 5. These are Anthropic’s own figures and the company notes that benchmark margins are a less reliable guide to real-world differences. Pricing is $4 per million input tokens and $20 per million output tokens, with a fast mode in Claude Code and the Claude Platform at $8 and $40.

A release digest for 2 to 8 November lists Claude Code versions 2.1.317, 2.1.318 and 2.1.319 without dates. Items include: auto mode now asks or denies when the server cannot verify a call, a fix for a sandbox escape when creating a worktree, a fix for a Bash permission bypass involving crafted environment prefixes, and a prompt before a web or shell tool call sends your account email over the network. Anthropic’s newsroom also lists the donation of Agent Skills to the Agentic AI Foundation on 22 October.

OpenAI

A third-party timeline of OpenAI release notes lists, among other items, Codex gaining faster steering for follow-up instructions (8 October) and Codex Security Cloud announced among DevDay developer tools (29 September). The 2 to 8 November digest says Codex CLI 0.167.0 adds an agent command center via a command-line option that marks unread results, plus a session picker for switching between backends while keeping drafts. We could not open OpenAI’s own pages, so treat these as second-hand.

What it means: security fixes in agent tools, such as a sandbox escape and a permission bypass, are a reminder to keep coding agents updated and to limit what they can run. See how to protect company data when using AI tools.

Cognition’s Devin

The same digest lists Devin changes: on 30 October, self-serve SSO with DNS-based domain verification, budget hierarchies with sub-limits, integrations for SonarQube, Linear and Vanta, and an approval-required default for enterprises without a personal access token policy; on 2 November, a cap of 1,000 repositories for multi-repo scans, up from 500; and on 6 November, admin controls for session sharing and a sign-out-everywhere option. These are the digest’s descriptions of Cognition’s release notes, which we did not open.

What it means: the emphasis is on administration and governance, which suggests vendors expect larger teams to be the buyers.

Microsoft Copilot Code

IT Brief reported that on 9 October 2026 Microsoft announced a Code capability within Copilot that 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. It reaches the Frontier programme at the end of October, according to the report. This is aimed at non-developers as well as developers, so it sits at the edge of our topic but affects who in a company can create software.

Timeline at a glance

Date (2026) Who Event Source type
14 August SpaceX / Cursor Acquisition of Anysphere completed (reported) News report
28 August OpenAI Announces plan to end Cursor model contract, proposing 12 November Timeline and report
22 September Anthropic Claude Opus 5.5 released Vendor announcement
20 October Cursor Agent SDK Vendor changelog
30 October Cognition Devin SSO and budget controls Release digest
3 to 4 November GitHub Copilot in JetBrains Air and Xcode Vendor changelog
5 to 6 November Cursor Live apps and voice chat Vendor changelog
9 to 10 November GitHub Claude Fable 5.2 and Gemini 4 Carbon in Copilot Vendor changelog

What it means for tool buyers

  • Ask which models you actually get. Supply deals and plan rules decide which models are available in each tool. The Cursor and OpenAI dispute shows this can change quickly.
  • Treat autonomous agents as a security question. Cloud agents that deploy apps or run shell commands need limits on secrets, network access and approvals. Several of the November fixes in Claude Code concern permissions and sandboxes.
  • Compare on your codebase. Vendor benchmarks are the vendors’ own. Use our guide to comparing AI tools side by side and run the same task in each tool.
  • Review generated code. Our guide on checking AI output for mistakes applies to code as much as to prose.
  • Understand the pricing unit. Seats, tokens, requests and credits all appear in this market. See how to read an AI tool’s pricing page.

What to watch next

These are questions drawn from the sources, not predictions.

  • Whether OpenAI and Cursor agree a different date or terms than the proposed 12 November cutoff.
  • Whether GitHub’s documentation clarifies which plans include the newest Claude and Gemini models.
  • Whether independent evaluations compare the coding agents on the same tasks.
  • How vendors handle security issues in agent sandboxes and permissions after the November Claude Code fixes.

Frequently asked questions

Is Cursor shutting down?

No, according to the report we read. It says Cursor is not shutting down and offers models from several providers.

Who owns Cursor now?

The report says SpaceX completed its acquisition of Anysphere, Cursor’s parent, on 14 August 2026. Confirm with an official announcement.

Which models are in GitHub Copilot?

GitHub’s changelog lists Claude Fable 5.2 (9 November) and Gemini 4 Carbon (10 November). We did not read plan details.

What did Cursor add recently?

Its changelog lists an Agent SDK on 20 October, live app building on 5 November and voice chat with agents on 6 November.

Is Claude Opus 5.5 the best coding model?

Anthropic says it leads on agentic coding benchmarks. That is Anthropic’s claim, and we have not tested it.

Which coding tool should I choose?

We cannot say, because we have not tested them. Compare tools on your own tasks, code and security rules.

Conclusion

The verified changes in AI coding tools this quarter point to more autonomy, more model choice and more attention to enterprise controls, alongside a business dispute that shows how dependent tools are on their model suppliers. Before you switch, read how to choose an AI tool, check how to protect company data, and use our trial checklist.

Sources

Keep reading