Factory AI pricing: Pro, Plus, Max, and team costs

Factory AI prices at a glance

Factory does not currently list a permanent free plan. Pro is therefore the practical evaluation tier. The product surface is broad even at Pro: the terminal agent, desktop application, SDK, background agents, local and cloud modes, billing statistics, and readiness dashboard are included. Plus and Max chiefly buy more Standard Usage, with managed computers appearing at Plus.

The relative usage language matters. “Approximately 5x” and “approximately 10x” describe plan relationships, not a guaranteed number of prompts, tokens, sessions, or completed tasks. One repository question and one autonomous migration consume very different amounts. Model selection, context length, tool output, retries, cache behavior, and computer compute all affect capacity.

What Pro includes for $20

Pro is the complete individual entry product. Droid can run interactively through the CLI and Desktop, operate as a background agent locally or in Factory’s cloud path, and run programmatically through the SDK or droid exec. The plan includes model access, usage and billing statistics, and the agent-readiness dashboard. Factory’s product also supports integrations and model selection, subject to account and organization configuration.

• Interactive work: terminal and desktop sessions for repository questions, planning, editing, tests, and review.
• Headless work: droid exec for scripts, CI, structured output, stdin pipelines, resumable sessions, and explicit exit codes.
• Local and cloud background agents: work can continue outside the foreground terminal, within the execution modes available to the account.
• Model choice: select supported models rather than buying a harness locked to one model family.
• Usage visibility: check usage in settings or with /limits, including fallback configuration after Standard Usage.

Pro does not include Factory-managed Droid Computers according to the current plan table. Local work and background agent access still exist. The distinction is persistent managed compute: Plus and Max can provision a long-lived Factory computer that retains packages, files, services, and configuration across sessions.

Pro makes sense for an individual trial, active local use, headless tasks on existing infrastructure, and teams that have not yet justified persistent vendor-managed computers. It can also establish a capacity baseline. If one developer uses Pro for a representative week, the observed work can be scaled more responsibly than a marketing multiplier.

What Plus buys for $100

Plus costs five times Pro and advertises approximately five times Standard Usage. The headline capacity increase is therefore roughly linear with the price. The differentiated feature is Droid Computers. Factory can provision a persistent cloud machine, or a user can target registered customer-owned machines through the broader computer system.

A managed computer is valuable when setup is expensive, work must continue after a laptop closes, services need to remain running, or several sessions return to the same prepared environment. Factory’s current computer documentation describes managed machines with 4 CPUs, 8 GB RAM, 6 GB swap, persistent state, and idle pause or resume behavior. Exact resources and availability can change, so confirm them for the intended workload.

The financial test is the incremental $80 per month. If a Droid Computer saves more than a small amount of setup or blocked time, the feature can cover the gap. If the upgrade is purchased only for usage, compare the effective accepted-task capacity with Pro plus Extra Usage. Prepaid overflow may be cheaper for occasional spikes; Plus is easier for steady demand.

What Max buys for $200

Max doubles the Plus price and advertises approximately twice its Standard Usage, or about ten times Pro. It retains Droid Computers and adds early access to features. This is the heavy individual tier for developers whose regular task volume consumes the Plus windows or whose work benefits from early product capabilities.

Early access is not a stable financial benefit by itself. Preview features can change, add rollout risk, or never become important to the buyer. Value the plan against task capacity first. A developer who consistently converts the extra agent work into reviewed, accepted code can justify $200 easily. A developer who opens more sessions than the team can review has purchased queue growth.

Max does not replace Extra Usage conceptually. Standard Usage can still run out, and the same Droid Core or Extra Usage choices apply. A long Mission can also pause at a limit. The larger allowance lowers the frequency; it does not turn the plan into uncapped autonomous compute.

How the three rolling limits work

Factory applies Standard Usage limits across 5-hour, weekly, and monthly windows. Every input and output token used by Droid counts, along with compute usage for Droid Computers. Standard Usage is always consumed before a fallback pool. The shortest window controls bursts, while weekly and monthly windows prevent continuous heavy work from behaving like an unlimited plan.

When a limit is reached, the current request finishes and later requests receive a token rate-limit error in the CLI or web interface. A long-running autonomous agent pauses at the same boundary. The user can run /limits to inspect status and choose Droid Core or Extra Usage, then retry without opening a new session.

Factory does not publish fixed quantities for these windows on the current documentation page. That makes direct arithmetic against another product’s documented tokens difficult. It also means community prompt counts are unreliable: the token shape of each task differs, and limits may evolve. Treat screenshots and forum anecdotes as observations about one account at one moment.

Droid Core: the included fallback pool

Droid Core is a Factory-selected set of leading open-weight models. While Standard Usage remains, Core model use counts against it. After Standard Usage is exhausted, the same Core models can draw from a separate token-rate-limit pool at no additional charge. Factory can change the model list as new options appear, and the current list is visible in the model settings.

This is a useful failure mode. A developer can continue routine work after the main frontier-model pool closes without immediately buying credits. The quality and fit of the fallback depend on the task. Mechanical edits, search, documentation, focused tests, and well-specified changes may work well. Architecture decisions, ambiguous debugging, or complex cross-system reasoning may justify waiting or enabling Extra Usage for a stronger model.

• Configure fallback before a deadline. Learn which Core models are available and how sessions behave when the main window closes.
• Route task types deliberately. Spend frontier capacity where errors and long trajectories are most expensive.
• Re-run verification. A fallback model operates through the same Droid harness, while its reasoning and tool behavior can differ materially.
• Record accepted-task yield. Free continuation has value only when its output survives review.

Droid Core also changes plan comparisons. “Stopped at the limit” is incomplete because Factory supplies a second included pool. “Unlimited” is also inaccurate because Core has its own rate limits. The correct model is two included pools with different model sets, followed by optional prepaid overflow.

Extra Usage: prepaid overflow

Extra Usage is a prepaid, US dollar-denominated credit balance. The minimum purchase is $10. Credits do not expire. Once enabled, the setting remains active until the user turns it off or the balance reaches zero. Standard Usage is consumed first, so credits are not spent while an included window still has capacity.

When Standard Usage is exhausted, the selected fallback determines whether subsequent work uses Droid Core or Extra Usage. Extra Usage provides continued access to models at Factory’s per-token rates. Current documentation directs users to live settings for model rates rather than presenting one universal rate, which is sensible because the models have different economics.

The prepaid design protects against an unbounded invoice if automatic top-up is absent. It can still hide task economics when several long agents spend from the same balance. Set alerts, assign ownership, and tag work to repositories or teams. A balance that vanishes during a Mission is operational data, not merely a billing surprise.

Missions, BYOK, and compute charges

Missions draw from the same rolling limits as ordinary sessions. Factory requires Extra Usage to be enabled before a Mission starts. If Standard Usage ends during the Mission, it can continue according to the configured fallback and available balance, or pause on a rate-limit error. Factory recommends Droid Core where it can perform the work because multi-agent execution magnifies token consumption.

Bring Your Own Key is included on Individual, Teams, and Enterprise plans with an allowance of free BYOK usage. After that allowance, Factory says BYOK usage is charged according to the plan. This differs from a pure local harness where the software vendor never charges above provider API spend. Ask Factory how the allowance and post-allowance charge apply to the selected deployment, gateway, and model.

Droid Computer compute counts toward limits alongside model tokens. Persistent managed compute therefore has two economic dimensions: the Plus or Max entitlement and consumption inside the usage system. The cost may be well justified if it removes environment rebuilds or keeps high-value work moving. Include it explicitly when comparing a Factory run with an agent on an existing developer machine or CI runner.

Teams and Enterprise pricing

Factory does not publish Teams or Enterprise dollar amounts. Teams supports up to 150 members and includes custom usage limits, onboarding and support, SSO, SAML and SCIM provisioning, zero data retention, and basic administration for model selection, autonomy, model access, and organization deny lists. Factory states that Teams plans are not affected by individual rate-limit changes.

Enterprise adds unlimited members, dedicated compute with partitioned inference, audit logging and activity trails, agent-readiness improvement work, an automation cookbook, on-premises deployment options, encryption key controls, data residency, session retention, network policy, an account manager, a customer engineer, priority support, SLAs, and custom onboarding. Exact packaging belongs in the order form.

1. Give sales a real workload model. Include active developers, task types, expected concurrent agents, Missions, remote computers, and chosen models.
2. Request a priced pilot. Measure the same outcomes before negotiating an annual baseline.
3. Separate seat and consumption terms. Confirm overages, pooled usage, Core availability, BYOK charges, computer compute, and renewal adjustments.
4. Map control availability. Verify which identity, retention, audit, deployment, and policy features exist in the proposed tier and region.
5. Define exit mechanics. Ask about transcript, telemetry, configuration, custom Droid, and audit export plus data deletion.

A quoted enterprise plan should be compared with the organization’s full alternative, including internal platform engineering. A set of individual Claude Code subscriptions is cheaper on a seat table and does not automatically provide centralized gateways, cross-model policy, persistent compute, Mission orchestration, audit integration, or deployment support.

Factory AI versus Claude Code pricing

The public individual ladders align closely. Factory Pro is $20, Plus is $100, and Max is $200. Claude Pro is $20 month to month, with Max tiers at $100 and $200. Both use included usage rather than a fixed number of unmetered model tokens. Both offer ways to continue through additional spend or different access paths.

Sticker price therefore does little to decide the individual comparison. Factory offers a model-independent harness, computers, and Missions. Claude Code offers the direct Anthropic path, broad first-party Claude surfaces, and inclusion in an existing Claude subscription. The cheaper option is often the plan already paid, unless the other harness materially improves accepted-task yield.

How to choose a plan without guessing

Questions people ask

Keep reading