Tabnine is a private AI coding assistant for companies that cannot send code to public AI services. Tricentis, a software testing company, acquired it on July 30, 2026, and tabnine.com now redirects to Tricentis. There is no free or self-serve plan: the free Basic tier ended in April 2025, and Tabnine is sold only as Enterprise, as secure SaaS or a private installation in your VPC or data center, by quote. The last published list prices were $39 per user a month for Code Assistant and $59 for the Agentic Platform, billed annually. Development continues, with release 6.6.6 on September 28, 2026.
Reviewed by EdGrows·Updated Sep 29, 2026This analysis is based on documentation, public user reports, and vendor materials — not yet on our own hands-on testing. How we rate
Tabnine is one of the oldest AI coding tools, offering code completion years before GitHub Copilot existed. In 2026 it changed hands. On July 30, Tricentis, a software testing company, announced that it had acquired Tabnine, with its interest centered on Tabnine's Enterprise Context Engine: a continuously updated map of an organization's code, documentation, tickets, APIs and infrastructure, which Tricentis plans to build into its AI testing platform so testing agents understand the effect of a change before they act. Terms were not disclosed.
The visible consequence came quickly. tabnine.com now redirects to Tricentis's products page, and the old pricing page to a Tricentis contact form. Tabnine's documentation remains online under its own name, and so does development: the release notes list version 6.6.6 on September 28, 2026. What Tricentis has not published is a roadmap for Tabnine as a standalone coding assistant, which is the question to ask before committing to a multi-year contract.
Many pages that rank for "Tabnine pricing" still describe plans that no longer exist. Tabnine sunset its free Basic tier on April 2, 2025, replacing it briefly with a time-limited preview, and later removed its lower-priced Dev plan for individuals. By 2026 the product was enterprise-only.
The last list prices Tabnine published before the acquisition were $39 per user a month for Code Assistant and $59 for the Agentic Platform, both billed annually, with no monthly option. Private deployments were priced by contract. Today everything goes through sales: the documentation states that Tabnine Enterprise accounts can only be set up together with the Tabnine team, and there is no way to buy a single seat.
Tabnine's documentation on September 29, 2026. The "plans" link leads to a Tricentis contact form; there is no public price list.
The documentation describes Tabnine as "a fully private, organization-aware AI coding agent designed to run entirely within your environment." In practice it is a set of tools for engineering teams: code completions, Tabnine Chat for questions about your codebase, and Tabnine Agent, which plans and carries out multi-step coding tasks in the IDE or from the command line, with MCP server support for connecting other systems.
The command-line side is changing fast. In September 2026 Tabnine released a plugin for OpenCode, the open-source terminal agent, and the Tabnine CLI is now built on it, while the older CLI gets upgrade instructions. An OpenCode-based agent for the IDE is in beta. Administrators get a console for managing the team, choosing which models are allowed and reviewing agent usage analytics.
Tabnine runs its own models for completions and chat, trained only on permissively licensed open-source code and hosted by Tabnine. For Agent and Chat, users can also pick from third-party models their administrator approves. The September 2026 list included Claude Opus and Sonnet versions up to Claude 4.8 Opus, GPT-5.5, GPT-5.4 and GPT-5.3 Codex, Gemini 3.5 Pro and 3.1 Pro, Mistral's Devstral 2, MiniMax-M2.7, GLM-4.7 and Qwen 3 Coder.
The idea is to let teams switch models by project. Work where privacy and legal risk matter less can use a performance-first model, while sensitive projects can use Tabnine Protected, built for compliance. A provenance and attribution feature, now extended to the CLI, flags generated code that matches open-source code under restrictive licenses, to reduce IP liability.
On data, Tabnine's documentation for its SaaS version says it never stores your code, does not share it with third parties and does not train on it unless you opt for a private custom model. It lists SOC 2 and GDPR compliance.
Private installation is what Tabnine is really selling. Besides the secure SaaS version, Tabnine Enterprise can run in a company's own VPC or on-premises, so prompts, code and model inference stay inside its environment. Private installations can also connect some third-party models through private endpoints the company controls.
This is why the price comparison with Copilot misleads. For a bank, a defense contractor or a hospital group whose security team will not approve code leaving the network, the alternative to Tabnine is often no AI coding assistance at all. The per-seat price is also not the whole cost: a self-hosted deployment means paying for GPU servers or cloud capacity to run the models, plus the engineering time to operate them.
Tabnine suits regulated organizations with strict data residency or IP requirements, large engineering departments that want central control over which models developers use, and companies already working with Tricentis on testing. For them, the private deployment options, administrator controls and model governance justify a quote-based enterprise purchase.
It does not suit individual developers, who cannot buy it, or small teams, who will get more features for less from mainstream tools. The acquisition adds a new risk for everyone else: the owner's strategic interest is the context engine inside its testing platform, and the future of the stand-alone assistant is not spelled out.
Among mainstream tools, GitHub Copilot Business at $19 per user adds IP indemnity and organization controls, though code goes to GitHub's cloud. Cursor Enterprise adds SSO and repository, model and MCP access controls, and the enterprise plan behind Devin Desktop can deploy in a company's own VPC. For teams that want open-source tools pointed at models they host themselves, Cline and Aider work with any compatible endpoint, at the cost of building the governance yourself. Sourcegraph Cody targets the same large-codebase enterprises from the code-search side.
For the broader field of coding assistants, see our guide to the best AI coding tools.
Tricentis, a software testing company, which announced on July 30, 2026 that it had acquired Tabnine to use its context engine inside Tricentis's AI testing platform. Terms were not disclosed. Since then tabnine.com redirects to Tricentis's products page, while Tabnine's documentation, plugins and release notes continue under the Tabnine name.
No. Tabnine sunset its free Basic tier on April 2, 2025, and later removed its lower-priced Dev plan. Today it sells only Tabnine Enterprise, either as secure SaaS or as a private installation, and accounts can only be set up together with Tabnine's team. Individual developers have no way to buy it.
Pricing is by quote through sales. The last list prices Tabnine published before the acquisition were $39 per user a month for Code Assistant and $59 for the Agentic Platform, both billed annually, with private deployments priced separately. Self-hosted customers also pay for their own GPU infrastructure, which the per-seat price does not include.
Yes, and that is its main selling point. Besides the secure SaaS version, Tabnine Enterprise can be installed privately in your own VPC or on-premises, so code and prompts stay inside your environment. Private installations can also connect some third-party models through private endpoints that you control.
Not by default. Tabnine's documentation says it never stores your code, does not share it with third parties and does not train on it unless you choose a private custom model for your team. Its own models are trained only on open-source code with permissive licenses, and it is SOC 2 and GDPR compliant.
Tabnine's own private models power completions and chat, and Tabnine Agent and Chat can also use third-party models approved by your administrator, including Claude Opus and Sonnet versions, GPT-5.5 and GPT-5.3 Codex, Gemini 3.5 Pro, Devstral 2, GLM-4.7 and Qwen 3 Coder. For strict IP requirements, the Tabnine Protected model is built for compliance.
The current documentation covers VS Code, the JetBrains IDEs, Visual Studio 2022 and 2026, Eclipse and Cursor, plus the Tabnine CLI, which since September 2026 is built on the open-source OpenCode agent through a Tabnine plugin. Admins manage the team, models and usage analytics from a console.
Yes. Release 6.6.6 shipped on September 28, 2026, adding a status command to the new OpenCode-based Tabnine CLI and improvements to an OpenCode IDE agent in beta, four days after release 6.6.3 introduced the Tabnine plugin for OpenCode. Tricentis has not published a long-term roadmap for Tabnine as a standalone product, which buyers should ask about before signing multi-year contracts.
The fast issue tracker for software teams, now built for agents too: free up to 250 issues, Basic $10 and Business $16 per user a month billed yearly.
The AI code editor most developers compare against, now owned by SpaceX: Pro from $16 a month yearly, agents, Projects and its own Composer models.
A free terminal coding agent that commits every change to git. It still works with almost any model, but its releases have slowed to a trickle.