Cursor Tab, Supermaven, and Windsurf Cascade: The 2026 Inline Code Completion Race After the Anysphere Acquisition

Anysphere did not buy Supermaven for the talent. It bought the timeline. Eighteen months later, the Ai Code Completion category has narrowed to three surfaces — Cursor Tab, GitHub Copilot, Windsurf Cascade — and the question is no longer who has the fastest Fill In The Middle model. It is who controls the inference budget when Claude Opus 4.7 sits behind the keystroke.

The Completion Layer Just Became a Routing Layer

Thesis: Inline completion is no longer a product. It is a contract between the IDE surface and a frontier coding model — and the contracts are being rewritten in real time.

For two years, completion tools competed on latency and context window. That race is over. Supermaven’s sub-10ms inference and long-context architecture got absorbed into Cursor in November 2024, and the standalone product was sunset on Nov 30, 2025 (webpronews). The remaining surfaces — Cursor, GitHub Copilot, Windsurf — all ship fast inline suggestions. The differentiator moved up the stack.

What developers actually buy now is which backend model the completion layer routes to, and how aggressively the vendor rations that routing.

That is a different business.

Three Releases, One Direction

The evidence is stacked tight. In the last fifteen months, every one of the structural moves has been about controlling the model layer, not the IDE.

Anysphere acquired Supermaven on Nov 12, 2024 (TechCrunch). The standalone Supermaven service sunset on Nov 30, 2025 (webpronews). Cursor Tab inherited the sub-10ms tech and the long context — now 200K standard, up to 1M tokens in MAX mode (Morph).

Then the backends jumped. GPT-5.3-Codex shipped in early February 2026 at 85.0% on SWE-bench Verified (marc0.dev SWE-Bench leaderboard). Claude Opus 4.7 followed on April 16, 2026 at 87.6% on SWE-bench Verified, up from 80.8% on Opus 4.6 (Vellum). That is a six-point ceiling jump in three months.

And the pricing models cracked under the new economics. Windsurf replaced its credit system with daily and weekly usage quotas in March 2026 (nocode.mba). GitHub paused new sign-ups for Copilot Pro, Pro+, and Student plans in April 2026 and tightened individual-plan limits (GitHub community). Two of the three remaining product-layer leaders rationed access inside a single quarter.

That is not three vendors making three independent decisions.

That is the model layer choking the product layer.

Who Moves Up

Cursor sits at the top of this market because it owns its surface and locked in long-context completion before the rest of the industry shipped competitive answers. Cursor Pro lists at $20/month with unlimited Tab completions plus a $20 frontier-model credit (Cursor pricing page), with annual billing knocking that down roughly 20 percent (NxCode). The Tab is unlimited; the expensive frontier calls are metered. That is the right side of the cost curve.

GitHub Copilot is the volume play. The free tier still ships 2,000 completions and 50 premium requests per month with no credit card (GitHub Docs), and Pro lists at $10/month with unlimited completions (GitHub pricing). Distribution through GitHub plus a route to GPT, Claude, and Gemini backends means Copilot does not need to win on speed — it needs to be present.

Windsurf wins the unlimited-Tab argument. Inline autocomplete is unlimited on every plan including Free and never counts against quota (Windsurf Cascade page), and Pro is $15/month with 500 credits and Cascade access to premium models (Verdent Guides). The product is positioned for developers who feel the Cursor metering.

Frontier model providers come out of this stronger, not weaker. Claude Opus 4.7 lists at $5 input and $25 output per million tokens (llm-stats). When three completion surfaces all need access to the same model, the model owns the negotiation.

Who Gets Left Behind

Any inline completion tool without a backend deal is now downstream of a market it does not control.

Standalone completion plugins — the category that produced Supermaven, Kite, Tabnine’s original product — collapsed into one of two outcomes: get acquired by a surface owner or get locked out of frontier model access. There is no third option.

Tools that built on the credit-billing model are scrambling. Windsurf’s March 2026 switch from credits to daily and weekly quotas is not a UX upgrade. It is a margin defense — credits were predictable for buyers and unpredictable for the vendor when usage spiked into expensive backend calls. Daily quotas cap the downside.

GitHub’s April 2026 sign-up pause is a louder signal. Microsoft is not closing Copilot. It is rationing access while the unit economics on Pro and Pro+ get rewritten. That is what a vendor does when the underlying model cost compounds faster than subscription revenue.

If your inline completion stack still bills like it is 2024, the next pricing email is already drafted.

Compatibility note:

• Supermaven standalone: Service sunset on Nov 30, 2025. Standalone plugins and accounts are no longer maintained. Action: migrate to Cursor.
• Windsurf credits: As of March 2026, the credit system was replaced with daily and weekly usage quotas. Old guides describing “credits/month” are stale.
• GitHub Copilot signup: New sign-ups for Pro, Pro+, and Student paused in April 2026. Status is a moving target — confirm before relying on it.

What Happens Next

Base case (most likely): The three-surface market — Cursor, Copilot, Windsurf — holds through the rest of 2026. Pricing tightens further as Claude Opus 4.7 and the Codex line absorb more of the hard edits. Signal to watch: A second major surface owner moving from credits to quota-based billing within the next two quarters. Timeline: 6 to 9 months.

Bull case: A frontier model provider — OpenAI or Anthropic — ships its own first-party completion surface that does not require a third-party IDE wrapper, compressing the stack from three layers to two. Signal: A native completion experience announced directly by a model lab, not a partner. Timeline: 9 to 12 months.

Bear case: Frontier model prices climb faster than subscription revenue can absorb, forcing one of the three surface owners into a credit-based metered model for inline Tab itself — ending the “unlimited Tab” era. Signal: Any of the three vendors quietly capping Tab requests on their entry plan. Timeline: 12 months.

Frequently Asked Questions

Q: How did Cursor’s acquisition of Supermaven reshape the AI code completion market? A: It removed the leading standalone completion vendor and folded its long-context, sub-10ms tech into Cursor Tab. The standalone service sunset on Nov 30, 2025 (webpronews), collapsing the category into three surfaces and forcing remaining competitors to differentiate on backend access and pricing, not raw completion speed.

Q: What is the future of AI code completion in 2026 as Claude Opus 4.7 and GPT-5.3 Codex push SWE-bench past 0.87? A: Claude Opus 4.7 crossed the 0.87 line at 87.6% on SWE-bench Verified (Vellum); GPT-5.3-Codex sits below it at 85.0% (marc0.dev SWE-Bench leaderboard). Completion surfaces are turning into model routers — the product that wins is the one that meters frontier calls without metering the keystroke.

The Bottom Line

The inline completion category consolidated and the model layer took the leverage. Watch the pricing pages: the next round of quota changes will tell you which surface owner is feeling the squeeze first.