LM Studio (with Bionic agent)
Free desktop app for running open coding models privately, now with a local agent; skip it if you need frontier quality.
Should you use it?
Pick it if
People who want local, private AI coding help in a desktop app rather than the terminal.
Best for: People who want local, private AI in a desktop app, not the terminal (Run AI on your computer Top 20)
Skip it if
You need the best coding quality; free Claude (Sonnet 5.5) ranks far higher on WebDev, as does GPT-6.1 Sol on ChatGPT Plus.
You give up: App is closed source; slower than Ollama on small models in one test
The short version
- What this is
- A free desktop app for finding, downloading and running open AI models on your own computer; its Bionic agent (July 2026) can work on your files.
- Why it matters
- It gives you a private, offline coding assistant at no cost, without using the command line.
- Free, including the Bionic agent; paid plans only add hosted cloud models s4.
- Local chats and documents stay on your machine, with no telemetry, per its policy s5.
- Quality depends on the model: Qwen3.8 27B ranks 28th of 142 on arena.ai WebDev s3.
Good at
Our take
LM Studio is the friendliest of the free local runners: a desktop app with a model browser, and since July a Bionic agent that can work on files on your PC s4 s7. Its privacy policy is the clearest of the three local tools we reviewed: local chats and documents never leave your machine and there is no telemetry s5.
As with any local tool, the ceiling is the model your computer can hold. Qwen3.8 27B ranks 28th of 142 on WebDev s3, solid but well behind free cloud options. Speed is platform-dependent: in Mozilla.ai's test it trailed the other local servers on small models on a Mac and on a Steam Deck, which the authors put down to LM Studio's prebuilt runtime s1. If you plan to use it at work, check the licence terms first, because the pricing page does not cover that s4.
Score
A score from 0 to 100 per job, against everything available today. Overall 80 across 1 job, the average of the job scores.
New to this? What the scores and labels mean
- 0 to 100
- Our score for one job, like writing or coding, compared with everything else out today. 100 is the best available today. For AI models we build it from public data, such as Arena's blind votes and Epoch AI's independent tests; apps and tools are scored on our own checklist. The same tool can score differently for different jobs, and where two likely ranges overlap, we call it a tie.
- Our verdict
- Use it: a top choice for most people. Worth trying: good, and worth a go. Depends: good for some people or one kind of job. Wait: too early to tell, or not out for everyone yet. Skip: something else does the job better.
- First look
- Just released. We say what it is and whether to bother, and give it a score once public results exist.
- No score yet
- None of the public sources we use covers it yet, or the only evidence is from a source whose numbers we can't use, so we describe it in words instead.
- Best buy
- Nothing cheaper scores as well for that job.
- Best free
- The best score for that job that costs nothing, through a free plan or by running on your own computer.
- Pay as you go
- No monthly plan: you pay for each use, usually through a developer account.
- Checked
- The day we last confirmed the score and the price against the sources. Prices change often, so check the date.
- AI model
- The engine underneath, like GPT or Claude. You use it through an app or, if you code, directly.
- Open source
- Free to download and run on your own computer, so nothing leaves your machine.
Our checklist score, going by an independent test, a public leaderboard and the maker’s pages.
Our checklist
2 sources
Sources and notes
Independent testPublic leaderboardMaker’s pagesChecklist 8/10: does the job 1 (Mozilla.ai, 17 Sep 2026, LM Studio runtime 2.28.2 on a Mac Studio M4 Max, a Steam Deck and an NVIDIA L40S server: it trailed llama.cpp, llamafile and Ollama on small models on the Mac and on the Steam Deck, s1; speed only, and Mozilla.ai makes llamafile, one of the rivals tested); underlying quality 1 (its catalog carries Qwen3.8 27B, s2, arena.ai WebDev rank 28 of 142, qwen3.8-27b, s3); value 2 (free, Bionic agent included, s4); privacy 2 (messages, chats and documents never leave your system; no telemetry, s5); limits 2 (Bionic 1.1.8 released 9 Oct 2026, s6).
Alternatives
Better or cheaper for coding, from our scores, then the ones we'd also look at. The same kind of product comes first.
- ClaudeApp80FreeMuch stronger coding model for $0, if sending code to a cloud service is fine. Side by side
- Claude Opus 5.5BetterText model100$20 a monthScores 100 in coding, against 80 here; $20 a month. Side by side
- OllamaPlatform80FreeOpen source with a one-command setup and a local API. Side by side
- llama.cppLibraryNot reviewed yetThe engine underneath, and faster on small models in Mozilla.ai's test.
Value for money
Where it sits in coding: higher is a better score, further left is cheaper. Green dots are the Best buys.
Evidence
Everything behind the call and the scores, folded so the page stays short. Spotted a wrong price, score or link? Report a mistake.
How we scored this
The coding score comes from our own checklist of the product.
Independent testPublic leaderboardMaker’s pages
Reviews are drafted with help from AI models, including Anthropic's Claude. Model scores come only from outside leaderboards and independent evaluations, by a written rule.
No one paid for this review or gave us free access, and we earn nothing if you sign up. How we score
Sources 8
- s1Mozilla.ai: Benchmarking llama.cpp vs llamafile vs LM Studio vs Ollama, 17 Sep 2026
- s2LM Studio model catalog
- s3arena.ai Code Arena: WebDev leaderboard
- s4LM Studio pricing
- s5LM Studio app privacy policy (effective June 2026)
- s6LM Studio changelog
- s7GIGAZINE: LM Studio releases the Bionic agent app (17 Jul 2026)
- s8arena.ai Text: Coding leaderboard
Something here out of date or wrong? Report a mistake and we will check it against the source.
History 3 changes
Every change to a rating, the price or our call, with the reason. Scores are against what exists on the day, so they drift down as better things ship. Older entries come from before we moved to scores.
- Evidence swapped: the InventiveHQ test was first published in June, outside our 90-day window; Mozilla.ai's test (17 Sep 2026) replaces it. Mozilla.ai makes llamafile, a rival in that test, so it counts only as some evidence (job 1). Rating unchanged.
- New: Coding ratedReviewed: coding (Bionic 1.1.8).
- Audit: evidence now says the InventiveHQ measurement is speed only; removed the unsourced claim that Qwen3.8 27B is the best coding model that fits most computers. Rating unchanged.
Use it well 4 guides on this site
Pages on this site that help with the jobs above.