Review Machine · Master Review · AI coding tool
Aider 0.86.2
Free software, and a bill only the model provider sends
Across 23 sources read on 22 September 2026, Aider 0.86.2 is free Apache-2.0 terminal software that all six review write-ups we read recommend for its git commits and its model freedom, whose spending is metered by the model provider with no cap inside the tool, and whose newest release is seven months old while a community fork has kept releasing.
Licence Apache 2.0Released 12 February 2026, on PyPIRepository 49,105 stars, 4,984 forksFork cecli v1.6.0, 19 September 2026Read 22 September 2026Skip to the verdictRead as text
Verdict age
ProvisionalUsers in · press thinThe version named is from 12 February 2026, seven months old at the date read, and nothing newer has shipped. The six review write-ups run from 17 March to 26 June 2026, so they describe the same build; the editorial pieces are from July and December 2025 and describe earlier ones. The repository's last push was 22 May 2026, and the community fork released v1.6.0 on 19 September 2026, so the tool a reader installs still moves, though not from here.
- Early · editors only
- Provisional · editors in, owners thin
- Settled · owner reviews read over months
From 23 sources to one verdict
Seven levels, in the same order in every Master Review. Each line is the takeaway; open a level for the evidence under it, or stop when you have enough.
What we read
23 sources, 10 Apr 2024 – 22 Sep 2026. Editorial, forum and buyer tiers thin.
By tier, heaviest first
The repository, the licence file, the release history, the package index and the docs, read 22 September 2026.
Three pieces from established publications, and none of them a review of the current release: a walk-through of 26 July 2025, a round-up of 24 December 2025 and a harness benchmark of 27 August 2026.
One report, 7 March 2026.
The repository's own figures, one API call each on 22 September 2026. Attention and workload, not satisfaction.
Three Hacker News threads, April 2024 to November 2025, and one search of the site's API on 22 September 2026.
Six dated review write-ups, 17 March to 26 June 2026, and two pages about the community fork.
Date window
10 Apr 2024 to 22 Sep 2026
Editorial reviews 26 Jul 2025 – 27 Aug 2026 · News 7 Mar 2026The project's own pages, the fork and the API figures were read on 22 September 2026; the review write-ups run from 17 March to 26 June 2026 and the oldest editorial piece from 26 July 2025.
Couldn’t read, so not used
- aider.chat/docs/usage/costs.htmlHTTP 404, the page no longer exists
Nothing counted here rests on a page we could not read. No review list was paged and no site filter applied.
Read, then rejected
- Simon Willison link blog, 31 July 2024 · a first look at an earlier version, before every release named here
- Awesome Agents review, 27 February 2026 · covers the same build as the six write-ups counted, and says nothing they do not
- Ivern AI review · its own fifty-task suite is reported without a method or a date, so no figure from it is cited
- Faros AI, best coding agents, January 2026 · a vendor listicle; Aider appears in a paragraph, with nothing on the themes counted
- AIMadeTools, one week with Aider, 22 May 2026 · a diary; its bill is named there and nowhere else it can be checked
- GenAlpha i, Aider in practice · no byline and no date of its own on the page
What they measured
49,105 stars, 1,380 open issues and a release seven months old. One dated benchmark put Aider at 0.8 million tokens for ten tasks where Codex spent 15 million.
Repository, 22 September 2026
49,105starsagainst 4,984 forks
GitHub · read 22 Sep1,380open issuesbeside 503 open pull requests
GitHub · read 22 SepNewest release per channel
0.86.2PyPI12 February 2026
PyPI · 12 Feb0.86.0GitHub9 August 2025, the newest entry on the releases page
GitHub · 9 Aug 20251.6.0ceclithe fork, 19 September 2026
GitHub · read 22 SepTokens for ten tasks, one model
The New Stack · 27 AugPrompt floor at startup
700tokensAider in architect mode
The New Stack · 27 Aug26,000tokensthe widest harness in the same benchmark
The New Stack · 27 AugWhat the write-ups reported spending
$473 weeksabout $67 a monthOne writer's own bill
Agent Finder · 30 May$47.306 weeksabout $1.12 a day of active codingOne writer's own bill
DEV Community · 28 May$70-80a monthheavy daily useOne writer's own bill
DevTools Review · 17 Mar$100-200a montha premium model in both rolesOne writer's estimate
Codegen · 23 JunLitmus puts typical developers at $10 to $40 a month and heavy users at $50 to $80. Five figures from five writers, none of it a survey.
Where they agree
Git commits, any model and a repo map, in the README and in all six review write-ups. 49,105 stars say attention came to look.
One tick per editorial review, in order of publication:TN The New StackIW InfoWorldTN2 The New Stack
Every change is a git commit
0/3The README lists git integration; the write-ups describe reviewing, amending and undoing AI edits with ordinary git tools. T1Official Forum Other
Any model, including a local one
0/3The docs and the README describe connecting to most LLMs; the write-ups put the count past a hundred providers. T1Official Other
A local model leaves nothing to pay
0/3Five of the six write-ups name running a model on your own machine as the zero-cost end of the range. Other
A repo map instead of pasted files
0/3The README lists codebase mapping; the write-ups describe the tree-sitter map as what keeps edits coherent without sending whole files. T1Official Other
Architect mode splits planning from editing
1/3A stronger planner and a cheaper editor; one benchmark gave its architect mode the smallest startup floor of twelve harnesses. T2Editorial Other
GitHub's own figures
The repository page reports 49,105 stars and 4,984 forks, from one API call on 22 September 2026. T3Buyers
49105stars4984forks
GitHub's own counts, from one call each. A star is attention, not quality.
What keeps coming up against it
The bill, uncapped, in all six write-ups. The terminal surface in six and a developer thread. A release seven months old in the project's own figures.
The provider bill, uncapped
Well documented · no ceiling in the toolsix write-ups describe the software as free and the spending as metered by the model provider, with nothing inside the tool to cap it.
- Agent Finder $47 over three weeks of moderate use, about $67 a month
- DEV Community $47.30 across six weeks, about $1.12 a day
- DevTools Review $70 to $80 a month for heavy use
- Codegen $100 to $200 a month with a premium model in both roles
- Litmus typical developers at $10 to $40 a month, heavy users at $50 to $80
DevTools Review · 17 MarDEV Community · 28 MayAgent Finder · 30 MayPondero · 5 MayLitmus · 26 JunCodegen · 23 Jun
the licence file gives the software as Apache 2.0, and no page of the project's own sets a price, a credit or a spending limit.
GitHub · read 22 SepGitHub · read 22 Sep
none. No established publication we could read has reviewed the current release, and the newest editorial piece measured harnesses rather than Aider alone. Thin
The surface is a terminal
0/3No inline autocomplete, no visual diff, no editor plugin. Forum Other
Context is manual
0/3Choosing the files is the work; one write-up puts the point where reliability falls at around 25,000 tokens. Other
A dense history
0/3One commit per change is a loud log for a team that squashes before merging. Other
The project's own pace
0/3Newest GitHub release v0.86.0 of 9 August 2025, newest package 0.86.2 of 12 February 2026, last push 22 May 2026, the fork's newest release 19 September 2026. T1Official T3Buyers Forum Other
GitHub's open queue
1380open issues503open pull requests
The issue search API reports 1,380 open issues and 503 open pull requests, one call each on 22 September 2026. T3Buyers Attention, not quality
GitHub's own counts. Work outstanding, not satisfaction, and a count of issues is not a defect rate.
Where they split
Five places the evidence pulls two ways, starting with what the tool costs against what using it costs.
What the tool costs against what using it costs
one dated benchmark put Aider at 0.8 million tokens for ten tasks where Codex spent 15 million.
Against that: every write-up reports that nothing inside the tool caps the spend, and their own bills run from about $6 to $200 a month.
Both truethe tool is free and the per-token economy is real; what varies is the model, the repo and the month.
Whether the project is keeping up
listed Aider among the open-source agents solving a real problem on 7 March 2026, and benchmarked it on 27 August 2026.
Against that: the newest release is seven months old, 1,380 issues and 503 pull requests are open, and a community fork has kept releasing, most recently on 19 September 2026.
Sothe project is stable and the fork is where the recent releases are; no publication we could reach has reported either.
The one hands-on against the review sites
concluded it would not recommend Aider against newer terminal agents, and described confusion over which directory owned the repository.
Against that: all six recommend it for terminal-first developers, and several call it the best free option.
Our readthe walk-through measures a terminal agent against newer ones and the write-ups measure free software against subscriptions; nobody we could reach has run the second test on the current release.
Stars against the queue
49,105 stars and 4,984 forks.
Against that: 1,380 open issues, 503 open pull requests and a newest entry on the releases page from August 2025.
Our readattention is not quality and an issue count is not a defect rate.
Who it's for, who should pass
For a terminal, a git history and a model of your own choosing. Pass if you want autocomplete, or a month that costs the same whatever you type.
It suits you if
- You already work in a terminal and in git.All six write-ups describe the same loop: describe the change, let the model edit, review the commit.
- You want to choose the model, or run one yourself.The docs describe connecting to most LLMs, the write-ups put the count past a hundred providers, and five of the six name local models as the zero-cost end.
- You want an audit trail of every AI edit.Each change is its own commit; two write-ups note the history is dense enough to squash before a merge.
- You want to try a frontier model without a subscription.The software is free and the bill belongs to the provider, so a quiet week costs almost nothing.
Pass if
- You want autocomplete inside your editor.All six write-ups name the missing inline completion as the trade.
- You need the month to cost the same whatever you type.Nothing in the tool caps the spend, and the bills reported run from about $6 to $200 a month.
- You need someone accountable for the bill or the code.One write-up notes there is no commercial support desk and no certification for enterprise procurement.
This is a reading of published reviews, not medical, financial or legal advice. For a decision about your health, your money or your rights, a qualified professional is the right next step, and not a review.
The verdict
Free software, a metered bill, and a release that is seven months old.
Aider is free Apache-2.0 terminal software that six review write-ups recommend for git commits and model freedom, whose spending is metered by the model provider with no cap inside the tool, and whose newest release is seven months old while a community fork has kept shipping.
Confidence, by tier
The licence file, the package index, the release history and the API figures agree on what the software is and when it last shipped.
Six dated review write-ups, 17 March to 26 June 2026, all describing the same build.
Three pieces from established publications, none of them a review of the current release.
Three threads, the newest from November 2025.
The repository's own figures: attention and workload, not satisfaction.
- Editorial evidence
- up to 13 months old
- Newest report
- 7 Mar 2026
- Read
- 22 Sep · month 40
It costs nothing to install, and however much you type.
Rests onSix review write-ups read on 22 September 2026 describe the same shape: free software under Apache 2.0, spending metered by the model provider, and nothing inside the tool that caps it. Their own reported bills run from about $6 to $200 a month, depending on the model.
Sources
23 sources, heaviest tier first. Every figure above comes from one of these.
- T1OfficialReleases page, newest entry v0.86.09 Aug 2025
- T1Officialaider-chat 0.86.212 Feb
- T1OfficialAider repository and READMEread 22 Sep
- T1OfficialLICENSE.txt, Apache 2.0read 22 Sep
- T1OfficialHISTORY.md release historyread 22 Sep
- T1OfficialConnecting to LLMsread 22 Sep
- T2EditorialDeveloper Walk-Through of Aider26 Jul 2025
- T2EditorialAI power tools24 Dec 2025
- T2EditorialAider, Claude Code and OpenClaw on one model27 Aug
- T2NewsOpen-source coding agents like OpenCode, Cline and Aider7 Mar
- T3BuyersGitHub REST API figuresstars, forks, open issues and open pull requests, one call eachread 22 Sep
- ForumAider: AI pair programming in your terminal, 432 points10 Apr 2024
- ForumWasting Inferences with Aider, 139 points13 Apr 2025
- ForumThe new Aider-CE fork is now official16 Nov 2025
- ForumHacker News search API, one call9,071 stories matchingread 22 Sep
- OtherThe new Aider-CE fork is now official16 Nov 2025
- OtherAider review, six months of use, updated 14 Sep 202617 Mar
- OtherAider's git-native workflow5 May
- OtherAider review, by pickuma28 May
- OtherAider review, three weeks of use30 May
- OtherAider review: terminal AI coding agent23 Jun
- OtherAider review26 Jun
- Othercecli, the community fork, v1.6.0 of 19 September 2026read 22 Sep
MethodOn 22 September 2026 we read six of the project's own pages, one set of GitHub REST API figures, three pieces from established publications, one news report, four Hacker News threads and searches, six dated review write-ups and two pages about the community fork, by web search and plain page fetch, and synthesised them with AI; we ran nothing and tested nothing.
Dates without a year are 2026.
The review as text
Aider is the terminal pair-programming tool whose current release is 0.86.2, published to PyPI on 12 February 2026 and licensed under Apache 2.0. We read 23 sources about it on 22 September 2026: six of the project's own pages, one set of GitHub REST API figures, three pieces from established publications, one news report, four Hacker News threads and searches, six dated review write-ups and two pages about the community fork that has kept releasing.
The price is the argument, in two halves. The software costs nothing and asks for no account; the spending is metered by whichever model provider you point it at, with nothing inside the tool that caps it. Every write-up we read describes that shape. The other thing to weigh is pace: the newest release is seven months old, 1,380 issues stand open, and a fork has taken over most of the releases.
Consensus
Aider earns its place across the sources we read for the same two reasons. Every change lands as its own git commit, so a session is reviewable and reversible with ordinary git tools, which the README, all six review write-ups and the 2024 launch thread all say. It connects to almost any model, local ones included, which the docs and all six write-ups confirm.
The reading is strong on what the tool is and thin on what it costs a reader. No source we could reach counted what Aider users spend, so every figure here is one writer's own bill, running from about $6 a month to about $200. No established publication we could read has reviewed the current release either; the newest editorial piece, of 27 August 2026, measured agent harnesses rather than Aider alone.
Recurring strengths
Every change is a commit. The README lists git integration, and all six write-ups describe reviewing, amending or undoing AI edits with ordinary git tools.
Any model you can reach, a local one included. The docs and the README describe it, all six write-ups name model freedom as the reason to choose Aider over a bundled agent, and five of the six point at a local model for a marginal cost of electricity.
A repo map instead of pasted files. The README lists codebase mapping, and five of the six write-ups describe the tree-sitter map as what keeps edits coherent in a large repository without sending whole files with every request.
Architect mode. Four of the six write-ups and one dated benchmark describe splitting the work between a stronger planner and a cheaper editor. That benchmark gave Aider's architect mode the smallest startup floor of its twelve harnesses, around 700 tokens against 26,000 for the widest.
Recurring complaints
The bill is real, uncapped and yours to watch. All six write-ups make the point in their own figures: $47 over three weeks of moderate use, about $67 a month; $47.30 across six weeks; $70 to $80 a month for heavy use; a warning that a premium model in both roles can reach $100 to $200 a month; and, from a fifth, typical developers at $10 to $40 with heavy users at $50 to $80. None of it is a survey, and the tool sets no ceiling.
The surface is a terminal. All six write-ups name the missing inline autocomplete, the absent visual diff and the lack of an editor plugin, and a developer in the 2024 thread wanted that integration too.
Context is manual. Four of the six describe choosing the files as the work, and one puts the point where reliability falls at around 25,000 tokens of combined context.
A dense history. Two write-ups note that one commit per change is a loud log for a team that squashes before merging.
The project's own pace. Six sources document it: the newest GitHub release is v0.86.0 of 9 August 2025 and the newest package 0.86.2 of 12 February 2026, the last push was 22 May 2026, 1,380 issues and 503 pull requests stand open, and the fork's newest release is v1.6.0 of 19 September 2026.
Where reviewers split
What the tool costs against what using it costs. A dated benchmark put Aider at 0.8 million tokens for ten tasks where Codex spent 15 million, so per token it sits on the thrifty side. Every write-up reports the other half: nothing inside the tool caps the spend, and the bills run from about $6 to $200 a month. Both are true, and neither says what your own month would cost.
Whether the project is keeping up. The New Stack listed Aider among the open-source agents solving a real problem on 7 March 2026 and benchmarked it on 27 August 2026. The repository's figures and the fork's say what that coverage does not: the newest release is seven months old, and the releases people install now often come from elsewhere.
The one hands-on against the review sites. The New Stack's walk-through of 26 July 2025 concluded it would not recommend Aider against newer terminal agents, and described confusion over which directory owned the repository. All six write-ups, written between 17 March and 26 June 2026, recommend it for terminal-first developers. The walk-through measures a terminal agent against newer ones; the write-ups measure free software against subscriptions.
Stars against the queue. GitHub's figures on 22 September 2026 show 49,105 stars and 4,984 forks, 1,380 open issues and 503 open pull requests. Attention is not quality and an issue count is not a defect rate.
Who it suits
You already work in a terminal and in git. All six write-ups describe the same loop: describe the change, let the model edit, review the commit.
You want to choose the model, or run one yourself. The docs put the count past a hundred providers, and five of the six write-ups name a local model that costs nothing but electricity.
You want an audit trail of every AI edit. Each change is its own commit, though two write-ups note the history goes dense enough to squash before a merge.
You want to try a frontier model without a subscription. The bill belongs to the provider, so a quiet week costs almost nothing.
Who should pass
You want autocomplete inside your editor. All six write-ups name the missing inline completion as the trade.
You need the month to cost the same whatever you type. Nothing caps the spend, and the reported bills run from about $6 to $200 a month.
You need someone accountable for the bill or the code. One write-up notes there is no commercial support desk and no certification for enterprise procurement.
This is a reading of published reviews, not medical, financial or legal advice. For a decision about your health, your money or your rights, a qualified professional is the right next step, and not a review.
One verdict a week: the most useful Master Review we finished, the complaint that kept appearing, and who should skip it. Get the weekly verdict.
Sources
Official
- GitHub, Aider repository and README, official, read 22 September 2026
- GitHub, LICENSE.txt, Apache 2.0, official, read 22 September 2026
- GitHub, releases page, newest entry v0.86.0 of 9 August 2025, official
- GitHub, HISTORY.md release history, official, read 22 September 2026
- PyPI, aider-chat 0.86.2 of 12 February 2026, official
- aider.chat, Connecting to LLMs, official, read 22 September 2026
Usage figures
Editorial and news
- The New Stack, Developer Walk-Through of Aider, 26 July 2025
- InfoWorld, AI power tools, 24 December 2025
- The New Stack, harness token costs, 27 August 2026
- The New Stack, open-source coding agents, 7 March 2026
Developer threads
- Hacker News, Aider: AI pair programming in your terminal, 432 points, 10 April 2024
- Hacker News, Wasting Inferences with Aider, 139 points, 13 April 2025
- Hacker News, the new Aider-CE fork is now official, 16 November 2025
- Hacker News, search API, one call, read 22 September 2026
Review write-ups
- DevTools Review, Aider review, 17 March 2026
- DEV Community, Aider review by pickuma, 28 May 2026
- Agent Finder, Aider review, 30 May 2026
- Pondero, Aider's git-native workflow, 5 May 2026
- Litmus, Aider review, 26 June 2026
- Codegen, Aider review, 23 June 2026
The fork
- GitHub, cecli, the community fork, read 22 September 2026
- Circus Scientist, the new Aider-CE fork is now official, 16 November 2025
On 22 September 2026 we read six of the project's own pages, one set of GitHub REST API figures, three pieces from established publications, one news report, four Hacker News threads and searches, six dated review write-ups and two pages about the community fork, by web search and plain page fetch, and synthesised them with AI; we ran nothing and tested nothing.
One verdict a week.
Every week, the most useful Master Review we finished: what the internet agrees on, the complaint that kept appearing, and who should skip it.