AgentLog

← All updates

Cursor

Developer
Anysphere
Type
Coding Agent

An AI code editor from Anysphere. Its Projects feature adds a coordinator agent that plans long-running work like features or migrations, then delegates the coding to many subagents running in the cloud, checking in with you only when there's something to review.

Get Cursor ↗

Latest on Cursor

Cursor Oct 1, 2026New model

Cursor adds Z.ai's GLM 5.3 and a faster Flash version

On October 1, 2026, Cursor said GLM 5.3 and GLM 5.3 Flash are available in the editor. Cursor's model docs describe GLM 5.3 as Z.ai's open-weight coding model after GLM 5.2, with a context window of 1 million tokens and Low, High, and Max reasoning settings. Cursor says GLM 5.3 Max is the best-scoring open-weight model on CursorBench 4.0, and GLM 5.3 is billed at the same per-token rates as GLM 5.2.

Your move: GLM 5.3 Max leads Cursor's open-weight benchmark

Read more

On October 1, 2026, Cursor said GLM 5.3 and GLM 5.3 Flash are available in the editor. Cursor's model docs describe GLM 5.3 as Z.ai's open-weight coding model after GLM 5.2, with a context window of 1 million tokens and Low, High, and Max reasoning settings. Cursor says GLM 5.3 Max is the best-scoring open-weight model on CursorBench 4.0, and GLM 5.3 is billed at the same per-token rates as GLM 5.2.

Why it matters

If you use Cursor, both models can be added from Settings, then Models. GLM 5.3 draws from the Other Models allowance, so check that pool before you rely on it for a long task.

X
Cursor Sep 23, 2026New bots

Cursor adds bots that watch deployments and scan every PR for security bugs

Cursor launched two new automated bots on September 23, 2026. Rollouts attaches a monitor to a pull request and watches the change after it deploys, reporting back whether each environment looks healthy, shows a regression, or is inconclusive. Security Review checks every pull request for exploitable bugs like injection flaws, authentication bypasses, and leaked secrets. Both are available now on Teams and Enterprise plans, turned on from the automations tab, and Cursor is giving teams free usage credits to try Rollouts on real changes for the first 10 days.

Your move: Available now on Teams and Enterprise plans, opt in from the automations tab

Read more

Cursor launched two new automated bots on September 23, 2026. Rollouts attaches a monitor to a pull request and watches the change after it deploys, reporting back whether each environment looks healthy, shows a regression, or is inconclusive. Security Review checks every pull request for exploitable bugs like injection flaws, authentication bypasses, and leaked secrets. Both are available now on Teams and Enterprise plans, turned on from the automations tab, and Cursor is giving teams free usage credits to try Rollouts on real changes for the first 10 days.

Why it matters

If your team already uses Cursor for code review, these bots add automated checks after a PR merges and while it deploys, catching some issues before users see them. They're opt-in, so nothing changes unless you turn them on.

Cursor Blog
Cursor Sep 23, 2026Efficiency update

Cursor cuts agent token costs by 7% with no drop in quality

Cursor published a blog post on September 23, 2026 describing changes that cut its agents' token costs by about 7% with no drop in quality. The savings came from a shorter system prompt, loading tools only when a task needs them instead of listing every tool up front, better placement of prompt caching, and removing instructions that older models needed but current models don't. Cursor said cold-cache rates, when saved context can't be reused, dropped by about 20%.

Your move: A behind-the-scenes cost cut, nothing to change on your end

Read more

Cursor published a blog post on September 23, 2026 describing changes that cut its agents' token costs by about 7% with no drop in quality. The savings came from a shorter system prompt, loading tools only when a task needs them instead of listing every tool up front, better placement of prompt caching, and removing instructions that older models needed but current models don't. Cursor said cold-cache rates, when saved context can't be reused, dropped by about 20%.

No action needed. Logged for the record only.

Cursor Blog
Cursor Sep 21, 2026New model

Cursor adds xAI's Grok 4.7 coding model

xAI released Grok 4.7 on September 21, 2026, and Cursor added it to its model picker the same day, alongside GitHub Copilot and xAI's own products. xAI built Grok 4.7 as a larger model than Grok 4.6, trained for longer on harder tasks, and says it scores higher on coding benchmarks meant to test longer, multi-step work. It costs the same per token as Grok 4.6.

Your move: Same price as Grok 4.6, aimed at longer, harder coding tasks

Read more

xAI released Grok 4.7 on September 21, 2026, and Cursor added it to its model picker the same day, alongside GitHub Copilot and xAI's own products. xAI built Grok 4.7 as a larger model than Grok 4.6, trained for longer on harder tasks, and says it scores higher on coding benchmarks meant to test longer, multi-step work. It costs the same per token as Grok 4.6.

Why it matters

If you use Cursor for longer or harder coding tasks, including in Projects, Grok 4.7 is worth trying since it costs nothing extra. Some early users report rough edges, like the model not yet working correctly in every part of Cursor, so test it on a task before relying on it for something important.

Cursor docs
Cursor Sep 10, 2026Projects launch (beta)

Cursor launches Projects, with one coordinator directing many subagents

Cursor launched Projects in beta on September 10, 2026, rolling out to all users the same day. Instead of opening a new chat for every task, you talk to one coordinator agent in a thread that stays open for as long as the work takes. The coordinator doesn't write code itself: it plans the work, creates and manages as many subagents as needed, delegates implementation to them in parallel, and brings the results back for you to review. Projects run on a cloud machine, so work continues even after you close your laptop, and can subscribe to Slack channels, schedules, or pull requests to act on things it detects on its own. Cursor says its heaviest internal users of Projects merge six times as many pull requests as before.

Your move: In beta, rolling out to all users; no separate pricing announced

Read more

Cursor launched Projects in beta on September 10, 2026, rolling out to all users the same day. Instead of opening a new chat for every task, you talk to one coordinator agent in a thread that stays open for as long as the work takes. The coordinator doesn't write code itself: it plans the work, creates and manages as many subagents as needed, delegates implementation to them in parallel, and brings the results back for you to review. Projects run on a cloud machine, so work continues even after you close your laptop, and can subscribe to Slack channels, schedules, or pull requests to act on things it detects on its own. Cursor says its heaviest internal users of Projects merge six times as many pull requests as before.

Why it matters

This moves Cursor from a chat-per-task tool toward something closer to a standing engineering teammate: one agent you brief, that spins up and manages a fleet of others while you're away. Worth trying if you have larger, multi-day coding work like features or migrations that don't fit neatly into a single chat session.

Cursor blog
Cursor Sep 2, 2026Update

Cursor's cloud coding agents can now run on your own servers

Cursor added a way to run its cloud coding agents on servers a team already controls, instead of only on Cursor's own machines. It works with AWS Lambda, Cloudflare, Coder, Daytona, Modal, Namespace, Vercel, and E2B. Cursor still plans and directs the agent's work from its own cloud, but the actual code editing and command execution happens inside the customer's own network. The update also adds worker pools that scale up and down on their own, and lets a whole team share machines instead of each developer needing their own.

Your move: Useful for teams that can't let code leave their network

Read more

Cursor added a way to run its cloud coding agents on servers a team already controls, instead of only on Cursor's own machines. It works with AWS Lambda, Cloudflare, Coder, Daytona, Modal, Namespace, Vercel, and E2B. Cursor still plans and directs the agent's work from its own cloud, but the actual code editing and command execution happens inside the customer's own network. The update also adds worker pools that scale up and down on their own, and lets a whole team share machines instead of each developer needing their own.

Why it matters

Companies with rules against sending code or secrets to outside services now have one less reason to avoid Cursor's cloud agents. Teams in finance, healthcare, or other regulated industries can get the same automated coding agent without moving code off infrastructure they already run and control.

Cursor blog