Cursor Router Pricing and Auto Mode: Cost, Balance, Intelligence, and Team Controls

Short answer: if your workspace is eligible—currently Cursor Teams or Enterprise—use Auto Cost for routine work and the firmest token budget, Auto Balance as the day-to-day default, and Auto Intelligence for difficult, high-value agent tasks. Select a named model manually when model identity, reproducibility, a provider-specific feature, or a known rate matters more than automatic routing. A team admin should set an Auto default only after a representative usage test; start with soft enforcement, then consider a hard lock when standardization outweighs individual choice. Cursor's official Router documentation does not extend Router to Hobby, Pro, Pro+, or Ultra.

Cursor Router is not a separate subscription or an unlimited add-on. The Teams or Enterprise plan establishes access, while included usage, routed-model charges, on-demand usage, and contract terms determine the working bill. The practical budget therefore has two layers: the workspace seat or contract, then the model usage created by each task.

The access and billing boundary

Teams is the self-serve entry point. Cursor lists Standard at $40 per active user per month and Premium at $120 per active user per month, with Premium providing five times the Standard Agent limits. Router is on by default for Teams. Enterprise uses custom pricing, offers pooled usage and deeper controls, and requires an admin to enable Router. Do not buy an individual plan on the assumption that its Auto selector includes this Router feature.

Each paid Teams seat has monthly included usage allocated to that user across Cursor Models and third-party models. The allowance resets with the billing cycle and does not transfer between teammates. Cursor's Teams page says each seat includes at least $20 per month of agent usage per person; usage beyond the included amount can become on-demand spend, billed in arrears, when that option is enabled. Enterprise pooling or negotiated terms can change the operational boundary, so a contract should win over the public self-serve assumptions.

The seat fee and the usage meter solve different problems. A seat buys managed access, administration, SSO, privacy controls, and usage visibility. It does not turn every routed request into zero-cost traffic. Use Cursor Pricing for the canonical plan matrix, then use the dashboard and the applicable contract for the live allowance and overage position.

Decision table

The Cost rates above come from Cursor's Teams pricing documentation. Balance and Intelligence do not have one fixed Router rate: the routed model and actual tokens determine the charge. Cursor says those modes average about twice the cost of Cost and can reach roughly two to four times Cost depending on the selected mode, but that is planning guidance rather than a maximum or quote. Enterprise buyers should validate the same concepts against their contract.

How to choose among the Auto modes

Cost is the cleanest baseline when the team wants automatic routing without letting the underlying frontier price become the default. It uses Cursor's earlier Auto routing logic and a fixed token schedule independent of which model serves the request. That makes the rate easier to forecast, but the result still depends on the task, context, and evolving model pool. It also consumes usage; “bundled” describes the rate, not unlimited work.

Balance is the sensible general default for an eligible team. Cursor describes it as optimizing intelligence, speed, and cost, and its classifier chooses a model from the request type and complexity. Use it when the organization wants strong everyday quality but has not proved that its routine workload needs Intelligence. Track both completion quality and spend because two requests with similar prompts can still route differently as context changes.

Intelligence should be an explicit high-value route, not a prestige setting. It aims at the most capable models for harder tasks and may justify its cost for difficult repository changes, stubborn failures, or long-horizon work. If routine edits dominate the queue, Intelligence can pay frontier rates where Balance or Cost would have been sufficient. A useful policy is to require a clear task trigger rather than letting every new chat inherit it silently.

Manual selection is safer when the model itself is part of the requirement. Choose it for a validated model-specific workflow, data or retention policy, predictable benchmark, special context behavior, or debugging a regression. Router is data-driven and Cursor-managed: users choose the optimization mode but cannot hand-pick the model that serves an individual routed request, and the available pool changes as models ship.

What the routed charge includes

Auto Cost uses its own per-million-token schedule and is exempt from the Cursor Token Rate. Auto Balance and Auto Intelligence instead charge actual usage at the routed model's API rate. When either route selects a third-party model on Teams or Enterprise, Cursor adds its $0.25 per million token rate. Cursor says that charge applies to eligible input, output, and cached tokens and sits on top of model pricing for included, on-demand, and BYOK usage.

Included usage is a credit boundary, not a different rate card. Balance and Intelligence consume the allowance faster than Cost, according to the Router docs. When a Teams user exhausts included usage, on-demand billing can continue if enabled; Cursor says it is enabled by default for Teams and can be governed with team-wide spending limits. Enterprise adds per-member limits and pooled usage options, subject to the contract.

A team should therefore monitor more than the monthly total. Review spend per user, mode, task class, routed model when exposed, and whether expensive tails come from context-heavy agent runs. The Cursor usage limits guide explains the meter and dashboard signals, while Cursor Pro vs Teams vs usage pricing separates personal subscriptions from managed workspace budgets.

Manual models, BYOK, and SDK access

Bring-your-own provider keys are separate from a Cursor Router decision. Cursor lets users add supported OpenAI, Anthropic, Google, Azure OpenAI, or AWS Bedrock credentials for compatible chat models. The provider owns the inference bill, while Cursor says the Cursor Token Rate still applies to eligible third-party BYOK traffic on Teams and Enterprise. Tab completion continues to use Cursor's built-in models, and provider data handling replaces Cursor's Zero Data Retention promise for those BYOK requests.

A Cursor SDK key is different from a provider BYOK key. The SDK runs Cursor agent workflows and follows the same pricing, request pools, and Privacy Mode rules as the IDE and Cloud Agents. A user key bills the user's plan; a service-account key bills the owning team. Router is exposed in the SDK as the Auto smart model with a Cost, Balance, or Intelligence parameter, but it remains limited to Teams and Enterprise availability.

The SDK is not a standalone chat-completions or raw model-inference API. Cursor does not document a public raw Router endpoint for arbitrary model calls. Programmatic buyers should therefore treat SDK use as another Cursor agent surface under the same plan or team budget, not as an independent API price sheet. Confirm Router appears in the key's model catalog before a production workflow depends on it.

Admin controls and a safe rollout

Admins can enable Router by team, and Enterprise can scope it by organization group. They can decide which modes members may select, set a default, show or hide the routed model for Balance and Intelligence, and allow or block underlying models. Cursor warns that blocking too many models reduces routing quality and can disable Router; its current documentation requires Cursor Grok 4.5 as the price-efficient option that anchors routing.

Soft enforcement makes Auto the default for each new chat but still allows manual switching. Hard enforcement locks the model picker to Auto. Start soft when the goal is adoption and measurement. Use hard enforcement only when the organization has validated its mode, model allowlist, exception process, and spend controls, because a locked default converts a recommendation into policy.

A practical rollout has five checks:

1. Record a manual-model baseline for common and difficult tasks, including cost and successful completion.
2. Pilot Cost or Balance with a representative group, not only the heaviest or most enthusiastic users.
3. Set a team-wide on-demand limit before the pilot and inspect user-level spend during it.
4. Compare median results and expensive outliers by task class; do not rely only on average cost per request.
5. Keep manual selection available until Router quality, model-policy compatibility, and budget behavior meet the team's acceptance threshold.

How to interpret Cursor's savings claims

Cursor reports that Router was trained on more than 600,000 live requests and evaluated through online tests across millions of requests. The vendor says three high-volume early-access accounts saved 30% to 50% on Auto-routed requests versus pricing all of that traffic at Opus 4.8 API rates, without a measured quality decrease. Cursor also reports Auto Balance delivered higher user satisfaction than Opus 4.8 at about 36% lower cost, and Auto Intelligence approached Fable satisfaction at about 60% lower cost for teams while improving satisfaction by about 15% over Opus 4.8 at nearly the same cost.

Those figures are Cursor-reported results, not general guarantees. The comparisons use Cursor's traffic, satisfaction signals, model pool, prices, cache-aware routing, and selected enterprise cohorts. Your repository mix, prompt length, cache behavior, model restrictions, contract, and definition of success can produce a different outcome. Use the claims to justify a controlled trial, not to book a fixed savings percentage.

Measure cost per accepted task, repaired incident, or reviewed commit alongside quality and rework. Cost per token can improve while total spend rises if Router increases usage volume. Conversely, a more expensive Intelligence request may be economical when it prevents repeated failed attempts. The correct buyer metric connects spend to completed work.

Final buying check

Choose Teams when self-serve seats, centralized billing, usage analytics, privacy enforcement, SSO, and team controls fit the organization. Choose Enterprise when pooled usage, invoicing or purchase orders, SCIM, advanced security, per-member controls, or negotiated Router terms matter. Do not upgrade from an individual plan solely to chase a savings claim; upgrade because the workspace and governance features are required and the usage trial supports the economics.

Before committing, verify plan eligibility, seat type, included usage, on-demand status, spending limits, Auto Cost rates, Cursor Token Rate treatment, Enterprise overrides, allowed models, and whether Router is enabled for the relevant team or group. If the broader decision is product fit rather than routing economics, continue to Cursor vs GitHub Copilot before standardizing the editor across a team.