LM Studio Bionic

LM Studio spent years as the easiest way to download and chat with open models on your own machine. Bionic, released on 16 July 2026, is the company’s step beyond chat: a separate desktop app that turns those same local models into a working agent. It splits work into Code projects — inspecting a repo, searching it agentically, and proposing inline diffs — and Work projects, where it produces documents, PDFs, slide decks and spreadsheets inside a sandbox.

The distinguishing choice is where computation happens. By default everything runs locally on models you download in the app, such as GLM 5.2 or Kimi K2.7 Code, so no code or document ever leaves the machine, and voice transcription runs offline through Mistral’s Voxtral. When a task outgrows local hardware you can hand it to LM Studio Secure Cloud, which serves frontier open models from US inference under a Zero Data Retention policy.

Pricing follows the same split: the app, local inference, web search tooling and LM Link across up to five devices sit on a $0 plan, while cloud usage is pay-as-you-go per token — GLM 5.2 at $1.50 per million input and $4.50 per million output, Kimi K3 at $3.00 and $15.00. A subscription called Bionic Pass is listed as coming but has no published price, and the pay-as-you-go page is light on credit expiry and context limits, so budget-sensitive teams should read the terms before committing.