AI coding assistants are sold by the seat, the credit and the token, often all three on one invoice.
This guide ranks thirteen for a team buying on behalf of developers: what usage costs once agents run all day, whether you can bring your own model keys, who may train on your code, and who owns the vendor.
AuthorHannah ReiterSenior Analyst, Business Applications
Vendors can pay for visibility on this page. It never changes what an entry
says about a product, including the criticism, and we earn nothing when you click through to a
vendor. How that works.
In short
What AI coding assistants software does
AI coding assistants use large language models to complete, write, edit and review source code inside an editor, an IDE extension or a terminal, increasingly as agents running multi-step tasks.
In this order: setup effort, what it really costs, how your data comes back out, whether
you can leave, and who each AI coding assistants tool is built for. Why those five, and why there is no
score out of ten, is on the how we work page.
AWS-centred teams already using it who need time to plan a move
Anyone making a new multi-year tooling decision today
Country is where the vendor is headquartered or contracts from, which is a
different question from where your data is hosted. Where the two tell different stories, the
entry says so.
A GPL-licensed editor that treats the model as a swappable part
Ranked #1 of 13 in Best AI Coding Assistants in 2026.
Free tierOpen sourcePublished pricingNorth America
The buying question with Zed is whether your developers will switch editor, because the money side is simple: Personal costs nothing with your own keys, Pro adds unlimited edit predictions and a token allowance, and Business adds organisation-wide model policies per seat.
Zed says it keeps prompts and code context only when a user opts in, and its provider agreements prohibit training. Traffic sent with your own keys or an external agent falls under that provider's terms instead.
What stands out
Open source editor
Own API keys
External agents
Where it costs you
Adopting it means moving every developer to a new editor
Pro includes only a small monthly token allowance for Zed-hosted models
Right for
Teams that want one open editor and freedom to swap models and agents
Wrong for
Organisations standardised on JetBrains IDEs or Visual Studio
United StatesFree Personal plan; Pro and Business per seat per month, published
Apache-licensed coding agent that asks before every edit and command
Ranked #2 of 13 in Best AI Coding Assistants in 2026.
Free tierOpen sourceSelf-hostablePublished pricingNorth America
The extension and CLI cost nothing; you buy inference through Cline at cost or point it at Amazon Bedrock, Vertex, Azure OpenAI, an OpenAI-compatible endpoint or a local runtime such as Ollama.
With your own keys, Cline's privacy notice says prompts go straight to that provider and Cline does not collect them. What you give up is predictability: nobody caps a developer's token use unless you set that up, and governance sits behind a sales call.
What stands out
Open source
Own API keys
On-prem option
Where it costs you
Enterprise pricing is not published; the pricing page sends you to sales
SSO, role-based access and central billing sit only in the Enterprise plan
Right for
Security-conscious teams that already hold model contracts with a cloud provider
Wrong for
Buyers who want one fixed monthly figure per developer
United StatesFree extension and CLI with inference at cost or own keys; Enterprise quoted
Agent for web, desktop and terminal, billed at provider token prices
Ranked #3 of 13 in Best AI Coding Assistants in 2026.
Free tierPublished pricingNorth America
The Hobby plan includes every product feature; what you pay for is tokens at the provider's price and, separately, minutes on 'orbs', the machines Amp creates per thread with your code and tools set up. Individual buys a block of orb minutes.
Amp says neither it nor its subprocessors train on your data unless you opt in, and that on Enterprise workspaces training can never be enabled. The meter is the risk: consumption pricing does not cap a busy week.
What stands out
No token markup
CLI and apps
Pooled credits
Where it costs you
Hosted by the vendor only; thread data is stored in a multi-tenant Google Cloud project
Purchased credits expire twelve months after purchase
Right for
Teams that want agent work billed by consumption, with no seat count
Wrong for
Organisations that must keep code context inside their own network
United StatesFree Hobby plan; Individual subscription and pay-as-you-go credits, published
MIT-licensed agent for VS Code, JetBrains and the command line
Ranked #4 of 13 in Best AI Coding Assistants in 2026.
Free tierOpen sourcePublished pricingNorth America
Kilo Code splits the bill in two: a platform fee per user for the Teams features, and inference, which is billed separately. The gateway advertises exact provider rates, and an optional Kilo Pass buys monthly credits with a bonus.
Its CLI is a fork of OpenCode. Anaconda's announcement says existing products, plans and support continue unchanged, which describes today and guarantees nothing. Kilo's privacy policy is silent on training; that depends on the provider you route to.
What stands out
Open source
Zero markup gateway
Owned by Anaconda
Where it costs you
Acquired by Anaconda in July 2026, so product direction now sits with a platform vendor
SSO, SCIM and audit logs are reserved for the custom-priced Enterprise plan
Right for
Teams wanting an open agent with a published per-user management layer
Wrong for
Buyers who insist on an independent single-product vendor
United StatesFree extensions with usage billed at provider rates; Teams per user per month, published
Terminal pair programmer that commits each change to git
Ranked #5 of 13 in Best AI Coding Assistants in 2026.
Free tierOpen sourcePublished pricingElsewhere
Aider suits a team that wants to spend on tokens and nothing else. It connects to most hosted models and to local ones through Ollama or an OpenAI-compatible endpoint, and its documentation warns that weaker models give poor results.
Leaving is trivial: uninstall it and the commits stay. The cost is on the management side. Keys, budgets and provider terms are handled per developer, and nobody at the vendor is contractually answerable for anything.
What stands out
Open source
Terminal
Own API keys
Where it costs you
No team administration, SSO or central billing of any kind
The operator, Aider AI LLC, publishes no address or country on its site
Right for
Engineers comfortable in a terminal who want every AI edit in git history
Wrong for
Managers who need usage reporting and access controls
Not stated by the vendorFree; you pay your chosen model provider directly
Flat plan per organisation, with a service fee on model usage
Ranked #6 of 13 in Best AI Coding Assistants in 2026.
Published pricingNorth America
A plan's monthly amount equals its included usage, and that pool is shared by everyone on it; beyond it you top up as you go. Augment sells a VS Code extension, a JetBrains plugin, the Auggie CLI and Cosmos, which chains triggers and agents into delivery automations.
Its documentation does not name the models behind them. Against the pass-through tools above, the 40% fee is what Augment charges for its context engine.
What stands out
Flat org price
40% service fee
CLI and IDE
Where it costs you
A flat 40% service fee is added to model usage billed at list price
The included monthly usage is small once up to 50 seats share one plan
Right for
Small engineering teams that want one invoice and shared agent automations
Wrong for
Teams that already buy model capacity and want it at cost
United StatesFlat monthly plan per organisation plus usage top-ups, published
AI inside JetBrains IDEs, metered in credits worth one dollar
Ranked #7 of 13 in Best AI Coding Assistants in 2026.
Free tierPublished pricingEurope
JetBrains publishes what a credit buys: roughly ten chat code generations or forty in-editor generations per credit, so quotas are easy to size and easy to exhaust.
AI Free gives three credits per thirty days, the paid tiers more, and top-up credits stay valid for twelve months. The AI Enterprise quota was listed as under review when checked. The assistant has no life outside these IDEs, which ties the AI decision to the editor decision.
What stands out
JetBrains IDEs
AI credits
Local models
Where it costs you
It only serves JetBrains IDEs, and the AI Free tier excludes Android Studio and Community editions
Some features are unavailable when you use your own provider key or a local model
Right for
Organisations whose developers already work in JetBrains IDEs all day
Wrong for
Mixed-editor teams that want one assistant across every tool
NetherlandsFree tier with monthly credits; per user per month, published
Anthropic's coding agent for terminal, IDE, desktop and browser
Ranked #8 of 13 in Best AI Coding Assistants in 2026.
Published pricingNorth America
Team and Enterprise seats bundle Claude Code with the chat product, SSO and managed settings; routing through Bedrock, Google Cloud's Agent Platform or Foundry puts the bill on your cloud account and keeps inference there.
Commercial data is retained for thirty days by default. The tool runs on your machine in the terminal, VS Code and JetBrains, with cloud sessions in Anthropic-managed virtual machines as an option. A price or limit change has no in-product alternative.
What stands out
Terminal agent
Claude models only
Cloud provider routing
Where it costs you
It runs Anthropic's Claude models only, so there is no second supplier
Zero data retention is limited to qualified Enterprise accounts and is not part of the standard plan
Right for
Teams that have chosen Claude and want terminal, IDE and CI covered
Wrong for
Buyers who want to switch model vendor month to month
United StatesPer seat per month on Claude plans, published; or pay-as-you-go API usage
AI editor and agent, now a SpaceX company with in-house models
Ranked #9 of 13 in Best AI Coding Assistants in 2026.
Free tierPublished pricingNorth America
Read the billing page before signing: included usage runs out, on-demand usage continues at API rates, and the Cursor Token Rate applies to third-party models while Grok and Composer are exempt.
That tilts the economics towards the owner's models. Privacy mode gives zero data retention with all providers and should be enforced team-wide. Ownership by SpaceX is the main change since earlier reviews of this market.
What stands out
AI editor
Owned by SpaceX
Usage-based overage
Where it costs you
Teams and Enterprise pay an added per-token rate when using third-party models
With privacy mode off, Cursor may use code and prompts to train its models
Right for
Teams that want a finished AI editor and accept its owner's model priorities
Wrong for
Buyers who want the editor vendor neutral between model makers
United StatesFree Hobby plan; per user per month plus on-demand usage, published
Cognition's agent with a desktop IDE, CLI and cloud sessions
Ranked #10 of 13 in Best AI Coding Assistants in 2026.
Free tierPublished pricingNorth America
Devin is built for delegation: it splits work across parallel cloud agents, runs its own tests in virtual machines and takes requests from Slack or Teams. Devin Desktop is the IDE.
The Teams plan covers up to 200 users; Enterprise adds VPC deployment and SSO and promises never to train without written consent. Opting out on a paid plan also switches on zero data retention with model providers. Legacy Windsurf enterprise accounts still bill on their old credit model.
What stands out
Formerly Windsurf
Cloud agents
Quota based
Where it costs you
Customer data may be used for model training by default until someone opts out
Usage is governed by daily and weekly allowances that the pricing page does not quantify
Right for
Teams that want to delegate whole tasks to agents running in the cloud
Wrong for
Buyers who need fixed, stated usage limits before committing
United StatesFree plan; Teams base fee plus per seat per month, published
OpenAI's coding agent, bundled into ChatGPT plans across CLI and cloud
Ranked #11 of 13 in Best AI Coding Assistants in 2026.
Free tierPublished pricingElsewhere
If you already buy ChatGPT Business, Codex is part of that seat. The documentation says business data is not used for training by default and that cloud tasks run on OpenAI-managed infrastructure you do not host.
For CI pipelines Codex can instead run on an API key at API prices. Forecasting is the weak point: Plus and Business plans get a band of local messages per five-hour window that depends on the model, and heavier use means buying credits.
What stands out
Bundled with ChatGPT
CLI and cloud
Credit top-ups
Where it costs you
Message limits vary by plan and model, so capacity per developer is hard to forecast
OpenAI's own legal pages refused automated fetches, so its country could not be confirmed here
Right for
Companies already paying for ChatGPT Business that want coding included
Wrong for
Teams wanting a choice of model provider inside one agent
Not stated by the vendorFree plan with limited use; ChatGPT plans per user per month, published
The assistant built into GitHub, now metered in AI credits
Ranked #12 of 13 in Best AI Coding Assistants in 2026.
Free tierPublished pricingNorth America
Copilot's plans are now sized in AI credits: Business carries 1,900 per user per month and Enterprise 3,900, and the unpaid individual plan is capped at 2,000 inline suggestions a month.
It runs in VS Code, Visual Studio, Vim and Neovim, JetBrains IDEs, a CLI and on github.com. The plans page we read did not say what happens when an allowance runs out, so ask before you size seats.
What stands out
GitHub native
AI credits
Many IDEs
Where it costs you
Agent and review features are tied to repositories hosted on GitHub
On individual plans, interaction data may be used for training unless the user opts out
Right for
Organisations on GitHub Enterprise that want procurement and policy in one place
Wrong for
Teams hosting code elsewhere or avoiding a single-platform dependency
United StatesFree plan; per seat per month with AI credit allowance, published
AWS's assistant, with its IDE plugins scheduled to lose support
Ranked #13 of 13 in Best AI Coding Assistants in 2026.
Free tierPublished pricingNorth America
The product page carries the end-of-support notice for the IDE plugins and names Kiro as the place for agentic coding, chat and MCP support.
Kiro is an AWS service with an unpaid plan of 50 credits a month and paid plans per user, each with a credit allowance, and it runs as an IDE and a CLI. Existing Pro customers keep indemnity meanwhile. New buyers should price Kiro, not this.
What stands out
AWS console
End of support 2027
IP indemnity
Where it costs you
AWS will discontinue support for the IDE plugins on 30 April 2027
The Free tier is capped at 50 agentic requests a month and leaves data collection on unless you opt out
Right for
AWS-centred teams already using it who need time to plan a move
Wrong for
Anyone making a new multi-year tooling decision today
United StatesFree tier; Pro per user per month, published
AI coding assistants use large language models to complete, write, edit and review source code inside an editor, an IDE extension or a terminal, increasingly as agents running multi-step tasks. The differences that matter are rarely in the feature list, so this is
the order we would work through them.
01
Decide whether you need a published price
13 of the 13 tools here publish what they cost; the other 0 quote per organisation. The ones you can compare without a sales call: Zed, Cline, Amp, Kilo Code, Aider, Augment Code, JetBrains AI Assistant, Claude Code, Cursor, Devin, OpenAI Codex, GitHub Copilot, Amazon Q Developer.
02
Decide how much the jurisdiction matters
These 13 vendors are established in 3 countries across 3 regions (North America 10, Elsewhere 2, Europe 1). That decides whose disclosure law applies to what the vendor holds, wherever the servers are.
03
Consider whether you want the source
4 of these are open source: Zed, Cline, Kilo Code, Aider. Hosting one yourself trades a subscription for maintenance.
Seat, credit or token: how AI coding assistants really bill
Three pricing shapes hide behind the word 'plan' in this market. Seat products such as GitHub Copilot and Cursor publish a price per user, then attach an allowance and charge for what exceeds it. Pass-through products such as Cline, Aider and Amp take little or nothing for the tool and hand you the model provider's token bill; Amp states it adds no markup, while Augment Code adds a flat 40% service fee on model usage.
Credit products sit in between: JetBrains AI Assistant defines a credit as one US dollar and tells you roughly how many generations it buys. A developer who runs agents all day can burn a month's allowance in a week, so the headline seat price tells you the floor and says nothing about the ceiling.
Ask each vendor what happens at the moment a developer's allowance is used up.
Pilot with your three heaviest users and extrapolate from their spend, not the average.
Check whether unused credits or top-ups expire; Amp's and Augment Code's lapse after twelve months.
Whose model is it, and can you bring your own key?
The assistant and the model are separate purchases even when one invoice covers both. Claude Code runs Anthropic's models only and OpenAI Codex runs OpenAI's, which is fine until a price or limit changes. Zed, Cline, Kilo Code and Aider let you supply your own provider keys or a local model, so an existing Bedrock or Vertex contract can carry the traffic.
Ownership now shapes the menu as well: Cursor exempts its first-party Grok and Composer models from the token rate it charges Teams customers on third-party models. If your security team has approved one model provider, start from the tools that can be pointed at it, because adding a new subprocessor usually takes longer than the pilot.
List the model providers your organisation has already approved before shortlisting.
Confirm that own-key mode keeps the features you need; JetBrains AI Assistant says some are unavailable.
Ask whether the vendor's own models are priced differently from third-party ones.
What happens to your code: training defaults and where it runs
Training policy in this category depends on the plan, and the default is not always the safe one. Devin may use customer data for training unless a paid customer opts out, and on its Teams plan only an administrator can. Cursor bars training only when privacy mode is on. Claude Code and GitHub Copilot separate commercial plans, which are not trained on, from individual ones, where the user's setting decides.
Location is the second question. Cline offers VPC, on-prem and air-gapped deployment to Enterprise customers, and Aider runs in your own terminal with keys you supply, whereas Amp states it is hosted by the vendor only and keeps thread context in its Google Cloud project. Developers on personal plans are the usual leak.
Get the training default for your exact plan in writing, including who may change it.
Block personal accounts on company repositories, or pay for seats so nobody needs one.
Ask where prompts and code context are stored and for how long.
Editor, extension or CLI, and who owns the assistant next year
Form decides how much change you are asking of developers. Zed, Cursor and Devin Desktop are editors, so adoption means leaving the current one. Cline and Kilo Code are extensions for VS Code and JetBrains IDEs plus a CLI; Aider and Claude Code start in the terminal. Then look at the owner, because this market reorganised itself during 2026.
Cursor was acquired by SpaceX in August, Kilo Code by Anaconda in July, windsurf.com now forwards to Devin, and AWS has set an end-of-support date for the Amazon Q Developer IDE plugins. Tools whose code is public under a permissive licence, and whose configuration lives in your repository, survive that kind of change better than a closed editor with your team's habits inside it.
Prefer tools whose rules and settings are plain files in your repository.
Sign annual terms at most; ownership changed at several vendors within one year.
Test the assistant in the editor your developers already use before proposing a switch.
What goes wrong most often when buying AI coding assistants software
Approving a per-seat price without modelling on-demand usage. Cursor, GitHub Copilot and Devin all continue billing after the included allowance, each by a different unit.
Assuming the vendor never trains on your code. At Devin the default allows it, and at Cursor it depends on privacy mode being switched on for the whole team.
Choosing on last year's ownership. Cursor, Kilo Code and the former Windsurf all changed hands or names, and Amazon Q Developer's IDE plugins have an end date.
Letting developers expense personal subscriptions. Individual plans of Claude Code and GitHub Copilot carry different training terms from the commercial ones.
07
Frequently asked questions
8 answers
What is the best AI coding assistants in 2026?
Zed leads our ranking of 13. Zed is an editor first, with its code public under GPL-3.0, and the Personal plan costs nothing while you supply your own API keys.
Agents from other vendors plug in over the Agent Client Protocol and Zed does not bill for them. The price of that independence is a migration: your developers change editor, and Zed's own hosted tokens are a small monthly allowance.
Which AI coding assistants tools publish their pricing?
13 of the 13, with the pricing model each one publishes:
Zed: Free Personal plan; Pro and Business per seat per month, published.
Cline: Free extension and CLI with inference at cost or own keys; Enterprise quoted.
Amp: Free Hobby plan; Individual subscription and pay-as-you-go credits, published.
Kilo Code: Free extensions with usage billed at provider rates; Teams per user per month, published.
Aider: Free; you pay your chosen model provider directly.
Augment Code: Flat monthly plan per organisation plus usage top-ups, published.
JetBrains AI Assistant: Free tier with monthly credits; per user per month, published.
Claude Code: Per seat per month on Claude plans, published; or pay-as-you-go API usage.
Cursor: Free Hobby plan; per user per month plus on-demand usage, published.
Devin: Free plan; Teams base fee plus per seat per month, published.
OpenAI Codex: Free plan with limited use; ChatGPT plans per user per month, published.
GitHub Copilot: Free plan; per seat per month with AI credit allowance, published.
Amazon Q Developer: Free tier; Pro per user per month, published.
The other 0 quote per organisation.
Is there a free AI coding assistants tool?
Zed, Cline, Amp, Kilo Code, Aider, JetBrains AI Assistant, Cursor, Devin, OpenAI Codex, GitHub Copilot, Amazon Q Developer offer a free tier or a free self-hosted edition.
Where are these AI coding assistants vendors established?
In 3 countries across 3 regions: North America 10, Elsewhere 2, Europe 1.
Zed: United States.
Cline: United States.
Amp: United States.
Kilo Code: United States.
Aider: Not stated by the vendor.
Augment Code: United States.
JetBrains AI Assistant: Netherlands.
Claude Code: United States.
Cursor: United States.
Devin: United States.
OpenAI Codex: Not stated by the vendor.
GitHub Copilot: United States.
Amazon Q Developer: United States.
Which AI coding assistants tools are open source?
Zed, Cline, Kilo Code, Aider.
Which AI coding assistants tools can you host yourself?
Cline. The other 12 are hosted by the vendor only.
What should you use instead of Zed?
Cline and Amp are the next two on this page. Cline is for Security-conscious teams that already hold model contracts with a cloud provider; Amp is for teams that want agent work billed by consumption, with no seat count.
Who should not buy Zed?
Organisations standardised on JetBrains IDEs or Visual Studio. Adopting it means moving every developer to a new editor.
If your AI coding assistants product belongs among these 13, tell us what it does and who it is for. Inclusion is an editorial call; what a listing is and is not is set out under software advice.
