What are Zed Terminal Threads?

Zed Terminal Threads shipped in Zed 1.3.5 on May 20, 2026. They let you run any terminal-based AI coding agent — Claude Code, OpenAI Codex CLI, Amp, or any custom CLI agent — as a thread inside Zed's agent panel sidebar, alongside Zed's own built-in agent. You can switch between conversations and shell sessions from the same panel list. Each thread can use a different agent, so you can run Zed's native agent in one thread and Claude Code in another concurrently.

How is this different from Cursor 3 Agents Window?

Cursor 3 Agents Window (April 2026) runs multiple AI coding agents in parallel inside Cursor — primarily Cursor's own models (Composer 2.5, GPT-5.5, Claude Opus 4.7, Gemini 3.5 Flash) plus cloud workers. Zed Terminal Threads is editor-agnostic about which agent runs — it embeds any terminal-driven external agent (Claude Code, Codex, Amp) as a thread. Cursor is opinionated and tightly integrated. Zed is the meta-routing IDE — bring your own agent.

Can I run Claude Code inside Zed in May 2026?

Yes. As of Zed 1.3.5 (May 20, 2026), Claude Code runs as a Terminal Thread in Zed's agent panel. You get Zed's editor and Claude Code's terminal-native agent in the same window with shared context. Note: starting June 15, 2026, Anthropic splits Claude Code billing into a separate credit pool (Pro $20 / Max 5x $100 / Max 20x $200 at full API rates with no rollover). Heavy users running Claude Code in Zed should plan to switch to direct API billing.

Which should I use for parallel AI coding agents?

If you want best-in-class for a single ecosystem: Cursor 3 Agents Window is the most polished. If you want to mix-and-match agents from different vendors: Zed Terminal Threads is the most flexible. If you want a single agent with disciplined subagent orchestration: Claude Code's built-in subagent system inside any editor (or just the terminal). Many serious teams now run Zed as the meta-IDE precisely because it doesn't lock them to one agent stack.

Quick Answer

Zed Terminal Threads vs Cursor 3 Agents vs Claude Code (May 2026)

Published: May 28, 2026

Zed Terminal Threads vs Cursor 3 Agents vs Claude Code (May 2026)

Zed 1.3.5 shipped Terminal Threads on May 20, 2026. Tagline from the Zed team: “Run Claude Code, Amp, or any terminal agent as threads in Zed’s sidebar.” That makes Zed the meta-routing IDE — the editor you use when you want to run multiple agents from different vendors at once. Here’s how it compares to Cursor 3 Agents and Claude Code’s native subagents.

Last verified: May 28, 2026.

TL;DR

	Zed Terminal Threads	Cursor 3 Agents Window	Claude Code Subagents
Released	May 20, 2026 (Zed 1.3.5)	April 2, 2026 (Cursor 3)	Iterative, mature in May 2026
Shape	Multi-agent panel, mix vendors	Multi-agent panel, Cursor-native	Single agent + spawned children
Vendor lock-in	None — bring any agent	Cursor models primary, others routed	Anthropic-only
Editor	Zed (Rust, fast)	Cursor (VS Code fork)	None — terminal-native (use any editor)
Free tier	Yes, Zed is free	Cursor free tier	None (Claude Pro $20 minimum)
Best for	Power users mixing agents	Cursor-committed teams	Anthropic stack users

What Zed Terminal Threads actually does

Zed’s agent panel always supported Zed’s own built-in agent for chat-style interaction. Terminal Threads (1.3.5, May 20, 2026) extends the panel:

Each thread can host either a chat conversation or a terminal session
Terminal threads run any CLI-driven external agent (Claude Code, Codex CLI, Amp, custom scripts)
All threads share Zed’s editor context — the file you’re looking at, the project root, the selection
You can switch between threads without losing state

From Zed’s docs (zed.dev/docs/ai/parallel-agents): “Each thread can use a different agent, so you can run Zed’s built-in agent in one thread and an external agent like Claude Code or Codex in another.”

Why this matters: as of May 2026, there’s no single AI coding agent that’s best at everything. Claude Opus 4.7 wins SWE-bench. GPT-5.5 wins on certain reasoning tasks. Gemini 3.5 Flash wins on cost. Zed lets you keep all of them open in tabs and pick the right one per task — without leaving the editor.

What Cursor 3 Agents Window does

Cursor 3 (April 2, 2026) introduced the Agents Window — a dedicated panel for parallel agent execution. Key features:

Run multiple Cursor agents simultaneously, each on different tasks
Mix local (inline edits) and cloud (long-running) agents
Composer 2.5 (Cursor’s in-house model, Kimi K2 lineage, May 2026) as the default low-cost model
Frontier models (Claude Opus 4.7, GPT-5.5, Gemini 3.5 Flash) on tap
Tight integration with Cursor’s chat, inline edits, and codebase indexing

Cursor’s strength: opinionated polish. Everything is tuned to work well together. The downside: you’re inside Cursor’s ecosystem. Models, billing, and tooling are Cursor’s choice.

What Claude Code subagents do

Claude Code (Anthropic) has supported subagents since late 2025. The pattern:

Main Claude Code agent decomposes a task
Spawns subagents for specific parts (e.g., one for tests, one for docs, one for code)
Subagents run in parallel, results merge into the main agent’s context
All subagents are Claude (Opus 4.7 or Sonnet 4.6) — single vendor

Claude Code is editor-agnostic — you run it in any terminal, including inside Zed Terminal Threads, VS Code’s terminal, or a plain SSH session. The subagent orchestration is built into Claude Code itself.

Where each one wins

Zed Terminal Threads wins

Mixing agents from different vendors. Run Claude Code, Codex, and Amp side-by-side. Pick the right one per task.
Fast editor. Zed is Rust-native and substantially faster than VS Code / Cursor for large repos.
Free. Zed has no editor license cost. You only pay for agent usage.
Open ecosystem. Any new terminal agent that ships becomes available immediately as a thread.
Anti-lock-in. If Anthropic raises prices, switch to Codex without changing editors.

Cursor 3 Agents Window wins

Single-vendor polish. Everything is designed to work together.
Composer 2.5 default. Cheap fast model for routine work without needing to think about it.
Mature cloud agents. Cursor’s cloud agent infrastructure is more battle-tested than Zed’s external-agent embedding.
Codebase indexing. Cursor’s repo embeddings are among the best.
YC / startup default. If your team is on Cursor, sticking with Cursor reduces friction.

Claude Code subagents win

Single-vendor depth. When the agent and subagents share the same model and tool surface, coordination is cleaner.
No editor required. Pure CLI. SSH into a server, run Claude Code, ship a fix.
Subagent orchestration is genuinely good. Better than ad-hoc multi-agent setups in Zed or Cursor.
Anthropic stack alignment. If you’re already paying for Claude Max, Claude Code is the natural fit.

The meta question: which workflow shape?

In May 2026, AI coding work falls into three shapes:

Interactive pair-programming — fast inline edits, autocompletion, refactors. Cursor wins.
Backlog burndown / async agents — assign work, walk away, review PRs. Devin (Cognition) or Cursor cloud agents win.
Heterogeneous power-user workflow — different agents for different tasks, all in one editor. Zed Terminal Threads wins.

A growing number of senior engineers use all three editors depending on the task — Cursor for active dev, Devin for backlog, Zed for “I want to compare Claude Code vs Codex on this specific problem.” That’s not waste — that’s specialization.

Pricing reality check

	Editor cost	Agent cost
Zed	Free	Pay your agents directly (Anthropic, OpenAI, etc.)
Cursor	Pro $20/mo, Ultra $200/mo	Bundled at tier + overage
Claude Code	Free (just a terminal)	Claude Pro $20 / Max 5x $100 / Max 20x $200 (billing splits June 15, 2026)

For pure cost optimization: Zed + direct API keys to your chosen models is often cheapest. You get the editor free and you pay raw API rates with prompt caching, batch discounts, etc.

For pure simplicity: Cursor Pro $20 is the cleanest single-bill, single-vendor option.

What changed in May 2026

Two important shifts:

Zed positioned itself as the meta-IDE with Terminal Threads (May 20). Zed is no longer “the fast editor” — it’s “the editor where you orchestrate multiple agents.”
Anthropic’s June 15 billing split makes the cost difference between Cursor (single bundled bill) and direct-API-via-Zed more salient. Power users running heavy Claude Code through any editor will want direct API keys.

Verdict

You’re a power user mixing agents: Zed Terminal Threads.
You want polished single-vendor experience: Cursor 3 Agents Window.
You want pure terminal Anthropic-native workflow: Claude Code (works fine inside Zed or any terminal).
You’re on a team where consistency matters: Cursor — everyone speaks the same tool.
You’re optimizing for cost: Zed + direct API keys.

The May 2026 takeaway: Zed Terminal Threads turns the editor into a routing layer. That’s a genuinely new shape in AI coding tools, and it’s the right answer for anyone tired of being locked into one vendor’s view of which agent is best.

Sources: Zed Release Notes May 2026, Zed Agent Panel docs (zed.dev/docs/ai/agent-panel), Cursor 3 release notes (April 2, 2026), Cursor Composer 2.5 announcement (May 18, 2026), Anthropic billing change announcement (May 13, 2026), digitalapplied.com AI Agent Stack Decision Tree.