Zed shipped a quiet but consequential run of updates recently, and it says a lot about where the AI code editor fight is headed. On October 7, 2026, the Rust-based editor pushed version 1.23.2 to its stable channel, adding JSONL and NDJSON table previews. Earlier, according to Zed’s release notes, bring-your-own-key (BYOK) support arrived for Claude Opus 5.5. Together these changes mark the clearest sign yet that the code editor market has split into two camps: platforms racing to add their own AI models, and platforms racing to let developers plug in whichever model they already pay for.
For a market that was, until recently, dominated by a single question (which autocomplete tool is best), the stakes have shifted. The question now is who controls the AI agent layer sitting inside the editor, and whether developers get to choose their own model or get locked into one vendor’s roadmap. Zed’s BYOK push, Cursor’s move into longer-horizon “Projects,” and VS Code’s steady cadence of Copilot updates all point to a market trying to answer that question before a single winner locks in.
Don't miss new tech stories on Google
Add Tech Insider once in the Google app and our stories appear in your news suggestions.
What Zed Just Shipped
Zed 1.23.2, which shipped on October 7, 2026, adds the ability to preview JSONL and NDJSON files as tables, with the rendered view opening alongside the raw source file, and it brings BYOK support for GPT-6.1 Sol using an OpenAI API key. The table preview is a small quality-of-life feature on its own, and it lands just two weeks after Zed 1.21.0, which is the release doing the heavier lifting. That update introduced BYOK support for Claude Opus 5.5 with an Anthropic API key and for several OpenAI models, including GPT-6 Astra, Sol, and Luna, along with on-the-fly model selection for new subagents, automatic subagent compaction, and dynamically fetched OpenCode models.
Zed has already moved well past 1.23.2, which signals the team is iterating fast even by the standards of an industry used to weekly ships. According to Zed’s release notes, stable 1.24.2 arrived on October 14, 2026 with Git tag creation, terminal-tab dragging into the Agent Panel and lower memory and CPU use during long agent threads, and stable 1.25.2 followed on October 21, 2026, with the preview channel at 1.26.1. For developers already running Claude Opus 5.5 through the API, the appeal of BYOK in Zed is straightforward: use the model you already pay for, inside the editor you already run, without a second subscription on top. Anthropic says Opus 5.5 costs 40% less to run than Opus 5, which makes that bring-your-own-key route easier to justify.
How Zed’s Model and Provider Support Has Widened
Zed’s release notes show the model list growing beyond the two names in the headline. The editor added GPT-6 Astra, Claude Fable 5.1, and Claude Opus 5.5 to its Amazon Bedrock provider, which gives teams that buy models through AWS another route to the same models. Zed also added Grok 4.7 to the SuperGrok and xAI providers and made it their recommended default, and it added support for DeepSeek Flash 4.1.
The model side moved too. OpenAI introduced GPT-6.1 Sol on September 29, 2026 as an upgrade to GPT-6 Sol for agentic coding, computer use, and professional work, and said it offered near-Astra intelligence at one-fifth of Astra’s standard API input and output pricing. For a team bringing its own key, that pricing difference flows straight onto its own API bill. On the Anthropic side, Opus 5.5 is available on Claude for Pro, Max, Team, and Enterprise users, and fast mode for Opus 5.5 is available in Claude Code and on the Claude Platform with up to 2.5x faster speed at $8 per million input tokens and $40 per million output tokens.
Zed also made the editor itself lighter. It reduced its binary size by about 33% on macOS, 28% on Windows, and 7% on Linux, and cut the Linux release binary size by about 23% on x86_64. Teams comparing BYOK against Zed’s hosted option should note that Zed offers a 14-day free trial of Pro with $5 of GPT-6 Luna usage and unlimited accepted edit predictions, and that hosted usage on Pro bills at API list price +10%. Bringing your own key avoids that markup, while the hosted plan trades it for a single bill.
Why Bring-Your-Own-Key Is the Feature That Matters
BYOK sounds like a minor toggle in a settings menu, but it reframes the entire pricing argument around AI coding tools. Editors that bundle a model into a flat subscription, Cursor’s Pro and Teams tiers among them, ask developers to trust one vendor’s selection and pricing for every task. Editors that support BYOK let a team mix models: a cheaper model for routine edits, a frontier model for hard refactors, and whatever the team’s security policy already approved for data handling.
That distinction matters more now that enterprise buyers are asking harder questions about where code and prompts actually travel. A BYOK setup keeps API calls under contracts the buyer already negotiated, rather than routing through a new intermediary. Zed’s bet is that this flexibility, not a single flashy model partnership, is what convinces engineering leads to standardize on one editor across a team. It is also a bet that plays to Zed’s strengths: a lightweight, Rust-built editor that has spent its early years competing on raw speed rather than on hosting its own frontier model.
Cursor’s Answer: Rollouts, Security Review, and Projects
Cursor is not standing still while Zed courts the BYOK crowd. On September 23, 2026, Cursor launched two enterprise-only features, Rollouts and Security Review, alongside a third capability called Projects. Rollouts monitors each deployment change and reports its health broken down by environment, giving teams a single place to watch whether a release is behaving after it ships. Security Review scans pull requests for exploitable bugs before merge, a feature squarely aimed at the “we can’t let an AI agent write code unsupervised” objection that still slows AI-tool adoption inside regulated companies. Both features are gated to Cursor’s Teams and Enterprise plans, not the consumer-facing Pro tier.
Projects is the more ambitious piece. Rather than optimizing for single-file edits or short tasks, Cursor built Projects to preserve context across months of work, delegate pieces of a task to thousands of subagents, and keep performing recurring work without requiring a fresh prompt each time. That is a direct pitch at larger engineering efforts: multi-sprint feature builds, migrations, or standing up new applications from scratch, the kind of work that previously required a human to keep re-explaining context to an AI assistant every session.
Then, on October 6, 2026, Cursor extended its iOS app so developers can see and reply to agents running locally on their own machine, straight from their phone. Combined with Rollouts, Security Review, and Projects, that is four distinct product moves inside two weeks. Teams already running Cursor alongside Claude Code now have a noticeably different toolset than the one they had at the start of September.
VS Code Keeps Its Own Pace
Microsoft’s editor has not sat out the sprint either. VS Code 1.141 shipped October 7, 2026, following 1.140 on September 30, which itself followed a run of releases, 1.136 through 1.140, that carried most of September’s Copilot updates. The cadence is the real story here: where Zed and Cursor are shipping named features with press-worthy announcements, VS Code keeps iterating on a near-monthly release train that most developers experience as a routine update notification rather than a launch event. That steadiness is also why VS Code still anchors most enterprise standardization decisions. Teams that already standardized on the Copilot agent sandbox in VS Code 1.141 are unlikely to switch wholesale just because a smaller competitor shipped a flashier feature.
The Usage Numbers Behind the Headlines
A 2026 editor-usage comparison circulating this month puts VS Code at 70.9% usage among its surveyed developer sample, with Cursor at 14.9%, down from 17.9% in 2025. That figure should be read with some caution since it comes from a secondary summary rather than a primary survey publication, but directionally it matches what engineering leaders describe anecdotally: VS Code remains the default, AI-native editors are still a minority choice even among developers who use AI tools daily, and the AI-editor category is still working out whether it is a replacement for VS Code or a specialized tool that sits alongside it.
That gap also explains why Cursor is leaning so heavily into enterprise features rather than consumer growth. If the AI-native editor category tops out around 15% of developers in a VS Code-dominated market, the fastest path to revenue is deepening wallet share among the teams already paying, not chasing incremental new users one download at a time.
What the Developer Survey Data Shows
Stack Overflow’s 2025 Developer Survey remains the most cited dataset on how working engineers actually feel about AI tools, and it is worth revisiting against this week’s releases. According to the survey, 84% of developers use or plan to use AI tools in their development process, a figure that explains why every major editor is racing to add agent features rather than treating AI support as optional. Separately, Stack Overflow found that 51% of professional developers use AI tools daily, suggesting adoption has moved well past the experimentation phase for a majority of the workforce.
The agent-specific numbers are more mixed. Stack Overflow reported that 31% of developers currently use AI agents, while 38% do not plan to use them, a split that undercuts the idea that agentic coding has already won over the profession. Among developers who have used agents at work, the upside case holds up: 69% say they experienced an increase in productivity. But trust remains the real friction point. Only 29% of developers say they trust AI outputs to be accurate, which is precisely the gap that features like Cursor’s Security Review and Zed’s BYOK model flexibility are trying to close, each from a different angle, one through automated verification and the other through letting teams choose a model they already trust.
Release Timeline: Three Editors, Two Weeks
| Date | Editor | Release / Feature | What Changed |
|---|---|---|---|
| September 23, 2026 | Cursor | Rollouts, Security Review, Projects | Deployment health monitoring, PR bug scanning, long-horizon project context (Teams/Enterprise only) |
| September 30, 2026 | VS Code | Version 1.140 | Continuation of September’s Copilot agent updates |
| October 6, 2026 | Cursor | iOS app update | View and reply to local agents from the Cursor iOS app |
| October 7, 2026 | VS Code | Version 1.141 | Latest stable release in the Copilot agent sandbox line |
| October 7, 2026 | Zed | Version 1.23.2 (stable) | JSONL/NDJSON file preview rendered as tables |
| October 7, 2026 | Zed | Version 1.23.2 (stable channel) | Stable release |
| Prior to October 7 | Zed | Version 1.21.0 | BYOK for Claude Opus 5.5 and GPT-6 Astra, Sol and Luna, subagent model selection, automatic subagent compaction |
Feature-by-Feature: Zed vs Cursor vs VS Code
Lining the three editors up side by side shows how differently each is trying to win the same developer. Zed is betting on openness and model choice. Cursor is betting on depth for paying enterprise teams. VS Code is betting on being the default nobody has a strong reason to leave. Readers who already compared these three head-to-head in our VS Code vs Cursor vs Zed cold-start benchmark will recognize the pattern: Zed wins on speed and flexibility, Cursor wins on agent depth, and VS Code wins on sheer market gravity.
| Capability | Zed | Cursor | VS Code |
|---|---|---|---|
| Latest stable version | 1.25.2 (Oct 21, 2026) | Rolling updates, last major Oct 6, 2026 | 1.141 (Oct 7, 2026) |
| Bring-your-own-key model support | Yes, including Claude Opus 5.5 and GPT-6.1 Sol | Not the headline feature; model selection built into plans | Copilot-centric, expanding model choice via extensions |
| Long-horizon project/agent memory | Subagent compaction, on-the-fly model selection | Projects: context preserved over months, thousands of subagents | Agent sandbox features in 1.141 |
| Security/code review automation | Not a named feature in this release cycle | Security Review (Teams/Enterprise) | Copilot agent sandbox, broader ecosystem extensions |
| Deployment monitoring | Not offered | Rollouts (Teams/Enterprise) | Not a native feature |
| Reported 2026 usage share | Not broken out separately in the cited comparison | 14.9%, down from 17.9% in 2025 | 70.9% |
Historical Context: How We Got Here
It is worth remembering how recent all of this is. VS Code built its dominance over roughly a decade by being free, fast, and endlessly extensible, long before anyone described an editor as “AI-native.” Cursor’s rise started as a fork of VS Code with AI features bolted on, and it only became a household name among developers once agentic coding, not autocomplete, became the feature that mattered. Zed entered later still, built from scratch in Rust with a pitch centered on raw performance, and has spent the past year retrofitting AI agent support onto that performance-first foundation rather than starting from AI and working backward.
That history explains why each editor is making a different bet this month. VS Code already won the default-install war, so it can afford to iterate quietly. Cursor needed a way to justify enterprise pricing beyond “better autocomplete,” so it built governance and monitoring tools. Zed needed a way to differentiate from both without the engineering headcount to build its own frontier model, so it opened the door to whichever model a team already trusts. None of the three strategies is obviously wrong, which is exactly why the market has not consolidated around one winner yet.
Market Impact: What This Means for Engineering Budgets
For engineering leaders, the practical impact of this week’s releases is less about which editor is “best” and more about how AI tooling budgets get allocated for 2027 planning. BYOK support in Zed means a team already paying for Claude Opus 5.5 or GPT-6.1 Sol access does not need a second line item just to use that model inside the editor, whereas Zed’s hosted Pro usage bills at API list price +10%. Cursor’s Teams and Enterprise gating on Rollouts and Security Review pushes the opposite direction, bundling governance features into a premium tier that justifies a higher per-seat price. VS Code’s steady Copilot cadence keeps the floor price low enough that neither competitor can simply out-price Microsoft.
The net effect is a market segmenting by buyer type rather than consolidating around a single tool. Cost-sensitive teams and individual developers gravitate toward VS Code’s free tier plus whichever Copilot plan fits their budget. Teams that already have model contracts elsewhere look at Zed’s BYOK flexibility to avoid duplicate spend. Enterprises worried about agent governance pay for Cursor’s Teams and Enterprise tiers specifically for Rollouts and Security Review. That segmentation is good news for all three companies in the short term, since none of them need to win the whole market to keep growing, but it also means the editor war likely continues for years rather than resolving into a single dominant platform.
The Trust Gap Nobody Has Closed
The Stack Overflow figure worth sitting with longest is the 29% trust number. Every feature shipped this month, Zed’s model choice, Cursor’s Security Review, VS Code’s agent sandbox, is in some sense a response to that number. None of them solves it outright. BYOK lets a developer pick a model they trust more, but it does not verify the model’s output. Security Review catches some exploitable bugs, but a scanner is not the same as a human reviewer with full context. Agent sandboxes constrain what an AI agent can touch, but constraining blast radius is different from guaranteeing correctness.
That gap, between 69% reporting a productivity gain and only 29% trusting the output, is probably the single most important number in this entire story. It explains why Cursor built a governance feature instead of a flashier model partnership, and why enterprises are willing to pay a premium for anything that narrows that gap, even incrementally.
What Comes Next: Five Predictions
- Expect VS Code to add its own version of BYOK-style model flexibility to Copilot within the next two to three release cycles, since ceding that ground to Zed indefinitely would be an unusual move for Microsoft given its history of matching competitor features quickly.
- Cursor’s Rollouts and Security Review will likely expand beyond Teams and Enterprise into a mid-tier plan once the company has enough usage data to justify a lower price point, following the same path many SaaS governance features take from premium-only to broadly available.
- Zed’s usage share will probably grow among developers who already use Claude or GPT models through separate API accounts, but it is unlikely to meaningfully dent VS Code’s 70.9% usage lead in the next year, given how much institutional inertia already favors VS Code in large organizations.
- The 29% AI-output trust figure will likely rise only slowly, since each editor’s governance features address a different slice of the problem rather than a single root cause, meaning no single release is likely to move that number sharply in either direction.
- Expect more editors to frame long-horizon, multi-month project memory, the capability behind Cursor’s Projects, as a standard feature rather than a differentiator by the second half of 2027, following the usual pattern where a category-defining feature becomes table stakes within a year of its debut.
How Developers Should Approach This Update Cycle
For individual developers and small teams, the practical advice is simple: do not switch editors just because of a single feature announcement. Zed’s BYOK support is genuinely useful if a team already has model API spend elsewhere, but it is not a reason to abandon a VS Code setup that already works. Cursor’s Rollouts and Security Review are enterprise features gated behind paid tiers, so smaller teams evaluating Cursor should weigh whether the Pro tier alone still justifies the price without those additions.
Engineering leaders evaluating a standardization decision should instead look at the pattern behind these releases rather than any single one. The direction each editor is moving, open model choice for Zed, governance depth for Cursor, steady default-install dominance for VS Code, tells you more about long-term fit than this week’s specific feature list. Teams that have already built workflows around tools like Cursor hooks inside a VS Code extension setup or migrated from older editors using guides like our Atom-to-Zed migration walkthrough are often better served by extending that existing setup than ripping it out for a headline feature.
The Bigger Picture for the AI Coding Tools Market
Step back far enough and this week’s releases look less like three unrelated product updates and more like three companies converging on the same conclusion from different starting points: the editor itself is no longer the product. The product is whatever sits between the developer and the AI agent, deciding which model runs, how much context it remembers, and how much a human has to trust it before merging its output. VS Code, Cursor, and Zed are each building a different answer to that question, and the Stack Overflow data suggests the market has not decided which answer it prefers. With 31% of developers using agents today and 38% saying they have no plans to start, there is still a wide swath of the developer population these companies have not converted at all, which may be the real race underneath this week’s headlines.
Frequently Asked Questions
What does BYOK mean in the context of Zed’s update?
BYOK stands for bring-your-own-key. It lets developers connect their own API credentials for a model, such as Claude Opus 5.5 or GPT-6.1 Sol, directly to Zed rather than paying for a bundled model subscription through the editor itself.
Is Zed’s BYOK feature available to all users or only paid plans?
The research available does not specify a separate paid tier requirement for BYOK in Zed’s 1.21.0 release. Developers should check Zed’s own release documentation for the most current plan-specific details before assuming free availability.
What is the difference between Cursor’s Rollouts and Security Review features?
Rollouts monitors deployment changes and reports health broken down by environment after code ships. Security Review instead scans pull requests before merge to flag exploitable bugs. Both launched September 23, 2026, and both are limited to Cursor’s Teams and Enterprise plans.
What is Cursor Projects and how is it different from a normal coding session?
Projects is designed for work that spans months rather than a single session, such as feature builds, migrations, or standing up new applications. It preserves context over time and can delegate pieces of work to a large number of subagents without requiring a new prompt for recurring tasks.
Which code editor has the largest market share in 2026?
According to a 2026 usage comparison cited in this article, VS Code holds roughly 70.9% usage among the surveyed developer sample, with Cursor at 14.9%, down from 17.9% the prior year. That figure comes from a secondary summary rather than a primary survey release, so it should be treated as directional rather than exact.
Do most developers trust AI-generated code?
Not fully. Stack Overflow’s 2025 Developer Survey found that only 29% of developers trust AI outputs to be accurate, even though 69% of those who have used AI agents at work reported a productivity increase. That gap is a major reason editors are adding governance and review features alongside AI capabilities rather than AI features alone.
Will Visual Studio Code add its own BYOK support to compete with Zed?
Nothing in the current VS Code 1.141 release notes confirms a BYOK feature matching Zed’s. Microsoft has historically matched popular competitor features over subsequent release cycles, but any specific timeline would be speculation rather than a confirmed roadmap item.
Should my team switch from VS Code to Zed or Cursor because of these updates?
Not based on a single feature release. Teams already invested in a working VS Code or Cursor setup should evaluate whether BYOK flexibility, governance features, or project-memory tools solve a specific problem they currently have, rather than switching editors purely because of this week’s announcements.
