Verdict
Factory suits engineering organisations that want one governed agent across laptops, CI, chat tools and cloud machines, with strong compliance and free choice of model 4,9,18. The catch is opaque usage: plans meter work through rolling rate limits and model multipliers rather than published token counts, and Business and Enterprise pricing is quote-only 3,4,5.
Good fit if you are
- Enterprises that need SOC 2 Type II, ISO 42001, zero data retention and on-prem or airgapped deployment
- Teams that want to hand work to an agent from Slack, Linear, Jira or CI as well as the terminal
- Developers who want to switch between Claude, GPT, Gemini, Grok and cheaper open-weight models in one agent
Look elsewhere if you are
- Buyers who want a fixed, published token allowance on each plan
- Teams of more than 10 that want SSO and audit logs without a sales call
- Developers who want a free tier or trial before paying
Pros and cons
What we like
- The same Droid runs in a desktop app, the CLI, an SDK, IDEs and persistent cloud or bring-your-own Droid Computers 1,8,10.
- Fine-grained safety: four autonomy levels, allow/ask/block command rules and a kernel-enforced sandbox for files and network 6,7.
- Broad model choice across Anthropic, OpenAI, Google and xAI, plus a cheaper open-weight Droid Core pool and bring-your-own keys 3,5.
- Published SOC 2 Type II, ISO 27001 and ISO 42001 reports, with zero data retention on Business plans 4,18.
- Droid with GPT-5.3-Codex ranks 10th of 142 entries on the Terminal-Bench 2.0 leaderboard 24.
What to watch
- Usage runs on rolling 5-hour, 7-day and 30-day rate limits; Plus and Max promise only about 5x and 10x Pro usage 3.
- Missions need High autonomy and paid Extra Usage, and pause when they hit a rate limit 3,6.
- Shared delegation from Slack, Linear, Jira and Microsoft Teams is still in Private Preview 9.
- Teams tops out at 10 seats; SSO, audit logs and zero data retention need a quote-only Business or Enterprise plan 4.
- The independent review base is small, with 37 reviews on G2 25.
Key facts
| Made by | Factory |
|---|---|
| Pricing | $20/month Pro; Teams $60/month + $40 per seat |
| Cost per developer | $20 a month (Pro plan, billed monthly; heavy users may need Plus at $100 or prepaid Extra Usage) |
| Where it runs | Terminal, IDE, Cloud (background), Web app, GitHub, Slack |
| AI models | Claude, GPT, Gemini and Grok models plus an open-weight Droid Core pool (GLM, Kimi, DeepSeek, Qwen and others), or your own keys 3,5 |
| Works with | VS Code, Cursor, Windsurf, JetBrains, Zed, GitHub, GitLab, Slack, Linear, Jira, Microsoft Teams |
| Security reports | SOC 2 Type II, ISO 27001, ISO 42001 |
| Free trial | Not published |
What Factory (Droids) is
Factory builds Droids, coding agents meant to cover the whole software lifecycle: planning, building, reviewing, testing and shipping 1. The same agent runs in the Factory App, the Droid CLI, the web, Slack, Microsoft Teams, Jira, Linear and CI 1. On top of the core agent, Factory sells a 'software factory' that adds code review, security analysis, documentation, QA and incident response 23.
The company raised a $150 million Series C at a $1.5 billion valuation in April 2026 21, then $200 million at a $5 billion valuation in September 2026, taking total funding past $400 million 20,23. It names Nvidia, Adobe, Palo Alto Networks, T-Mobile and RBC among its customers 20.
What it can do
You start a session in the desktop app or terminal and pick a mode: Normal to execute, Spec to plan read-only first, or Mission for long multi-feature projects 6. Autonomy Level (Off, Low, Medium or High) decides what runs without approval, and permission rules can allow, ask for or block specific commands 6. An optional OS-level sandbox limits writes to the working directory and network traffic to allowed domains 7. MCP servers, skills, hooks and custom subagents extend the agent 12.
For background work, Droids run on persistent Droid Computers, either Factory-managed or your own machine 10. Teams can delegate from Slack, Linear, Jira and Microsoft Teams, and Droid answers with a session or pull request to review 9. A GitHub Action and a GitLab CI component run automated pull request reviews 11. If a session goes wrong, rewind restores the chat and your files to an earlier point 13.
- Actions. Droids read and edit code, run commands and tests, and open pull requests, including from a Slack, Linear or Jira request 9. Missions orchestrate long multi-feature projects, and Droid Exec runs headless in CI with a chosen autonomy level 1,6.
- Setup. Install the Factory App or the Droid CLI and start a session in any repository; editor integrations come from the VS Code Marketplace, JetBrains AI Agents or the Zed extension list 8.
- Reviewing its changes. Droid shows changes in IDE-native diffs, and delegated work arrives as a pull request from the Factory Droid GitHub App for a human to review and merge 8,9.
- Analytics. Droid exports OpenTelemetry metrics to your own collector, and a hosted Analytics API reports usage, credits and productivity 17.
Pricing
Individual plans cost $20 (Pro), $100 (Plus) and $200 (Max) a month 2. Each is governed by rolling 5-hour, 7-day and 30-day rate limits; Plus gives about five times Pro usage and Max about ten times, but Factory does not publish the token allowance behind them 3. Models draw down usage at different multipliers, from 0.04x for GPT-6 Luna to 4x for Claude Fable 5.1 5. When limits run out you can drop to the open-weight Droid Core pool at no extra cost, or buy prepaid Extra Usage from $10 that never expires 3. Missions only run with Extra Usage turned on 3. Teams costs $60 a month plus $40 per seat for up to 10 seats, so a five-person team pays $260 a month 2. SSO, audit logs and zero data retention start on the quote-only Business plan 4.
| Plan | Price | Includes |
|---|---|---|
| Pro | $20/month2,3 | Factory App, Droid CLI and SDK, cloud and local background agents, billing, usage and readiness dashboards |
| Plus | $100/month2,3 | Everything in Pro with about 5x Pro usage |
| Max | $200/month2,3 | Everything in Plus with about 10x Pro usage and early feature access |
| Teams | $60/month per team + $40/month per seat2,4 | Pro rate limits for each seat, up to 10 seats, centralized billing |
| Business | Quote only2,4 | Custom usage limits, SSO and SAML/SCIM, zero data retention, audit logging, admin controls for models, autonomy and network policy |
| Enterprise | Quote only2,4 | Unlimited seats, dedicated compute and partitioned inference, on-prem deployment, sub-organizations, data residency, priority SLAs |
| Extra Usage | Prepaid credits, $10 minimum3 | USD credit drawn down after plan limits run out; does not expire |
Security and data
Factory publishes SOC 2 Type I and Type II reports plus ISO 27001 and ISO 42001 certificates in its trust center 18. The Droid CLI runs locally and does not upload or index a copy of your repository; prompts and context go to the model providers or gateways you configure 15. With enterprise endpoints such as Azure OpenAI, Bedrock and Vertex, traffic goes straight to the provider under contracts that support zero retention, and hybrid or airgapped deployments keep model traffic and telemetry inside your network 15. Business plans add zero data retention and audit logging, and Enterprise adds data residency and on-prem deployment 4. Factory says it does not use your code as training data 19.
- Zero data retention. Zero data retention on Business and Enterprise plans 4
- Data residency. Data residency on Enterprise; cloud, hybrid, EU and airgapped deployment patterns 4,14
- Training on your data. Factory says it does not use customer code as training data 19
- AI models. Anthropic, OpenAI, Google and xAI plus open-weight models; customer keys, gateways and self-hosted models supported 5,15
What customers say
Independent reviews are thin. Factory AI holds 4.5 stars from 37 G2 reviews on an unclaimed listing, and one reviewer praises the time saved but still checks every output 25. On the independent Terminal-Bench 2.0 leaderboard, Droid with GPT-5.3-Codex ranks 10th of 142 entries at 77.3%, an unverified submission from February 2026 24. Factory names Nvidia, Adobe, EY, Palo Alto Networks, Adyen, T-Mobile and RBC as customers 20,21.
| Platform | Rating | Reviews | Seen |
|---|---|---|---|
| G2 (Factory AI)25 | 4.5 / 5 | 37 | 8 October 2026 |
Benchmarks
| Benchmark | Result | Source |
|---|---|---|
| Terminal-Bench 2.024 | 77.3% with GPT-5.3-Codex (rank 10 of 142; unverified submission, Feb 2026) | Independent leaderboard |
| Terminal-Bench 2.024 | 69.9% with Claude Opus 4.6 (rank 23 of 142; unverified submission, Feb 2026) | Independent leaderboard |
What Factory claims (not independently verified):
- Used by hundreds of thousands of developers 20
- Revenue doubled month over month for six straight months before the April 2026 Series C 21
- Sessions using Factory Router cost 63% less in aggregate than the same work at frontier-model rates 22
- Routed sessions reach 99% of Claude Opus 4.7's Terminal-Bench 2 pass rate 22
- Adyen: Factory Router saved thousands of dollars in inference costs in one week 22
Concerns raised by others:
How we scored it
| Criterion | Why | Score |
|---|---|---|
| What it can do 30% of the total | Droids plan, edit, run tests and open pull requests from the CLI, IDEs, Slack, Linear and CI, work locally or on cloud Droid Computers, and extend through MCP 6,9,10,12. | 5/5 |
| Pricing 20% of the total | Pro, Plus, Max and Teams prices are public, but usage tiers are described only as about 5x or 10x Pro and Business and Enterprise are quote-only 2,3,4. | 4/5 |
| Control 20% of the total | Autonomy levels with allow/ask/block rules, an OS-level sandbox, IDE diffs and PR review, a rewind that restores files, and org-wide maximum autonomy plus audit logs are all documented 6,7,8,13,16. | 5/5 |
| Security and data 15% of the total | SOC 2 Type II, ISO 27001 and ISO 42001 reports are published 18, zero data retention comes with Business plans 4, and Factory says customer code is not used for training 19. | 5/5 |
| Customer evidence 15% of the total | Droid placed 10th of 142 on Terminal-Bench 2.0 but as an unverified submission 24, G2 shows 4.5 from only 37 reviews 25, and named customers include Nvidia and Adobe 20. | 4/5 |
| Total | Weighted sum of the 5 levels, out of 10. | 9.3 |
Sources
- Factory: agent-native software development
- Pricing
- Individual Plans
- Organization Plans
- Available Models
- Autonomy Level
- Sandbox
- IDE Integrations
- Remote Delegations
- Droid Computers
- Automated Code Review
- Model Context Protocol (MCP)
- Full Changelog
- Factory for Enterprise
- Data Flows & Privacy
- Compliance & Audit
- Telemetry & Analytics
- Factory Trust Center
- Security
- Factory raises $200M at $5B valuation
- Series C
- Factory Router cuts inference costs by 63%
- Factory Raises $200M as It Builds Agents Across the Software Lifecycle
- Terminal-Bench 2.0 Leaderboard
- Factory AI Reviews
