Supermaven vs GitHub Copilot: Speed vs Breadth of Context

Supermaven no longer exists. It shut down in November 2025 after Anysphere (the company behind Cursor) acquired it a year earlier and folded its speed architecture into Cursor Tab. If you’re choosing a completion tool today, your real options are Cursor Tab or GitHub Copilot — not Supermaven.

But the Supermaven-vs-Copilot comparison still matters, because it explains exactly why Cursor Tab is as fast as it is, what GitHub Copilot traded away for breadth, and how the two tools represent genuinely different bets on what a completion tool should be.

Who this is for

Developers trying to understand the speed-vs-ecosystem tradeoff in AI code completion — either because you’re evaluating Cursor against Copilot today, or because you’re researching what made Supermaven worth acquiring. If you need a Cursor Tab vs Copilot decision guide for 2026, the framing here will transfer directly.

What Supermaven was

Supermaven launched in February 2024 with a single, loud claim: inline completions were slow, and it was going to fix that. The tool shipped with a 300,000-token context window at a time when GitHub Copilot’s inline completion context sat at around 8,192 tokens. That’s a 37× difference in how much code each tool could see when deciding what to suggest next.

By the time Supermaven 1.0 shipped — powered by a new model called Babble — the context window had expanded to 1,000,000 tokens. The accompanying blog post framed the comparison directly: “We’ve trained Babble, a new model, which is 2.5x larger than our previous model and expands its context window from 300,000 to 1 million tokens.”

The tool supported VS Code, JetBrains, and Neovim. It had no agentic features, no chat interface, no multi-file editing. It did one thing: watch your cursor and suggest the next line before you asked.

The context window gap

Context window is where the historical numbers are sharpest. At Supermaven’s February 2024 launch, the comparison was unambiguous:

Tool	Context window	Date
Supermaven	300,000 tokens	Feb 2024
Supermaven 1.0 (Babble)	1,000,000 tokens	~2024
GitHub Copilot (inline)	8,192 tokens	Early 2024
GitHub Copilot Chat	64,000 tokens	Dec 2024

Copilot expanded its Chat context to 64k tokens in December 2024, closing some of the gap for conversational interactions. What Copilot’s inline completion context window is in 2026 is not clearly documented — no verified primary source survives for a current figure, so treat any specific claim you read elsewhere as unconfirmed.

The practical implication of Supermaven’s 1M-token window: the tool could read your entire codebase as context, not just the file you had open. For large TypeScript monorepos or Go services with deep cross-package dependencies, that mattered. Copilot in early 2024 was often suggesting completions without seeing the function you were actually trying to call.

Speed — what we can and can’t claim

Supermaven’s launch post included a benchmark: 250ms for Supermaven, 783ms for GitHub Copilot, measured in some number of video frames. These figures did not survive adversarial verification — they’re vendor-produced marketing numbers from Supermaven’s own blog, with no reproducible methodology documented. Don’t cite them as ground truth.

What is accurate: Supermaven’s architecture was deliberately built around minimizing completion latency as the primary design constraint. Developer discourse — across Hacker News threads and Reddit comparisons through 2024 and into 2025 — consistently described it as faster than anything else available. “Feels instant” was a common descriptor. That community signal is qualitative, not a benchmark, but it was consistent enough to be meaningful.

GitHub Copilot’s team has published claims of latency improvements — “35% reduction” in one post — but those figures also failed independent verification. Copilot’s speed improved over 2024, but by how much, measured how, is not something a primary source will cleanly tell you.

The honest summary: Supermaven was faster. By how much, precisely, no external party measured.

Where Copilot won

Speed was the one thing Supermaven beat Copilot at. Copilot won everywhere else.

Agentic features. Supermaven had none. It was a completion tool, full stop. GitHub Copilot Agent Mode — which lets Copilot autonomously decide which files to change, propose terminal commands, and iterate until the task is complete — reached VS Code Stable in March 2025. JetBrains added agent mode support in 2025. NeoVim’s official extension still doesn’t support it.

IDE breadth. Copilot’s official feature matrix covers VS Code, JetBrains, Visual Studio, and others at varying capability levels. VS Code and JetBrains get the full stack: chat, agent mode, MCP integration, next-edit suggestions, and vision. NeoVim gets code completion via the official extension only, though the third-party CopilotChat.nvim plugin adds chat on top of that. Supermaven’s IDE support — VS Code, JetBrains, Neovim — was narrower and never expanded beyond standalone completion.

Enterprise trust. Copilot had GitHub’s distribution, Microsoft’s enterprise agreements, and IP indemnification baked into Business and Enterprise tiers. Supermaven was a startup with a few dozen enterprise customers. For organizations where legal review gates tool adoption, that asymmetry was decisive.

Pricing

Supermaven’s pricing didn’t survive the research verification pass — specific historical tier prices from their pricing page couldn’t be confirmed. The product is shut down, so this is moot for new users.

GitHub Copilot’s current tiers as of June 2026:

Tier	Price	What you get
Free	$0	2,000 completions + 50 chat/agent requests/month
Pro	$10/month	Unlimited completions, AI credits, model selection
Pro+	$39/month	Higher AI credits, Claude Opus and other premium models
Max	$100/month	Highest AI credit allowance, priority access to new models and features
Business	$19/user/month	Team features, policy controls, unlimited completions
Enterprise	$39/user/month	Advanced security, audit logs, Copilot knowledge bases

Note: GitHub moved from premium request units (PRUs) to token-consumption-based billing effective June 1, 2026. Code completions and Next Edit Suggestions remain included in all paid plans and don’t consume AI Credits. New sign-ups for Pro, Pro+, and Max are temporarily paused as of this writing; existing customers can upgrade between paid tiers.

What happened to Supermaven’s speed

Anysphere acquired Supermaven on November 12, 2024. The announcement framed it as a technology acquisition: the Babble model and the architecture behind its low-latency completions were going into Cursor. Supermaven officially shut down approximately one year later, around November 21, 2025.

Cursor Tab — Cursor’s native inline completion — is the direct heir to Supermaven’s speed work. If you want what Supermaven was, you want Cursor Tab. The product line is different (Cursor is a full fork of VS Code with its own subscription), but the speed philosophy carried over.

Verdict

In 2024 and early 2025, picking Supermaven over Copilot made sense if you were a solo developer or small team, your primary bottleneck was completion latency, and you didn’t need agentic features. The tool was genuinely faster and had a dramatically larger context window for inline work.

In 2026, that choice is no longer available. Your decision is between Cursor Tab and GitHub Copilot.

Pick Cursor if inline completion speed and context breadth are your top priority, you work in VS Code or want a VS Code-based IDE, and you can live with Cursor’s subscription pricing.

Pick Copilot if you need agent mode across your existing JetBrains or Visual Studio setup, you’re working in an enterprise environment where Copilot’s compliance story matters, or you want to stay in your current IDE without switching to a Cursor fork.

The speed gap that made Supermaven notable hasn’t disappeared — it’s that Cursor Tab is where it lives now.

For a detailed breakdown of how Cursor Tab and Copilot compare on benchmarks and agentic workflows today, see our Cursor vs Copilot 2026 comparison.

Caveats

This article covers Supermaven as a historical product. All Supermaven data comes from snapshots and blog posts made before the November 2025 shutdown — primary sources are still accessible at time of writing but may not remain so. GitHub Copilot’s current inline completion context window size is unconfirmed; the 64k figure applies to Copilot Chat (December 2024), not inline completions. No independent latency benchmarks comparing Supermaven and Copilot were produced by a neutral third party. Toolchew has no affiliate relationship with Cursor, GitHub, or any tool mentioned in this article.