AI coding assistants

An AI coding assistant writes and changes code alongside a developer: inline completions as you type, a chat that knows your codebase, and an agent mode that edits files, runs commands and, in some tools, opens pull requests from the cloud. Developers at startups, software companies and large engineering organizations use them, from one person on a free plan to teams that need central billing and security terms. They differ in where they run (inside an existing IDE, in their own editor, or in the terminal), how usage is counted (requests, credits in dollars, or multiples of another plan), which model providers they offer, and what the business terms say about training on your code, retention and IP indemnity.

We compare them on the same six indicators, read from each vendor's pricing, docs, trust and legal pages:

• Pricing clarity: whether the team plan has a published price, can be bought online and month to month, whether included usage is stated in requests or dollars, whether extra usage has a published rate, and whether the pages leave prices out or contradict each other.
• Free plan and trial: a permanent free plan with completion and chat, and whether it has unlimited completions, an agent mode and a training opt-out; otherwise the length of the trial and whether it needs a card.
• Price for 5 developers: the monthly cost for 5 developers on the cheapest organization plan with central billing, a no-training commitment and chat plus an agent mode with usage included, billed annually where offered, with the arithmetic shown; 0 when the pages cannot price it.
• Usage terms and model choice: usage stated in requests or dollars, a published overage rate, a spending cap admins can set, models from more than one provider, and unlimited inline completions.
• Code privacy and IP protection: no training on your code, zero or short data retention, IP indemnity for generated code, and content exclusion or a public-code filter.
• Compliance and admin: the vendor's own SOC 2 Type II and ISO 27001, SAML single sign-on, SCIM, an audit log and an EU data option.

Speed, suggestion quality and benchmark results depend on the model, the codebase and the developer, so we do not score them. Each review says which pages it was written from and when. Where we have run a product hands-on, the evidence is on its page.