Aider pricing

About

Aider is a free, open-source AI pair programmer that runs in your terminal. It edits code across your local git repository, automatically commits its changes with attribution, and builds a compact “repository map” so the model has context about your codebase. The project is published under an Apache-2.0 license, with the source on GitHub, and is maintained by Aider AI LLC. Aider notably writes roughly 70% of its own new code each release.

Critically, Aider has no product price — there is no subscription, no per-seat fee, and no enterprise tier. It is a bring-your-own-key (BYO) tool: it connects to a model provider using an API key you supply, so your only cost is the LLM API tokens you consume, billed directly by that provider (OpenAI, Anthropic, Google, DeepSeek, xAI, and many others). You can also run it at zero token cost using local models via Ollama, or free API tiers such as OpenRouter’s free models and Google’s Gemini 2.5 Pro Exp.

Despite carrying no price, Aider is one of the most popular open-source AI coding tools — roughly 40,000+ GitHub stars as of 2026 — and competes head-to-head with paid, hosted assistants like Cursor, GitHub Copilot, and Claude Code. Its differentiation is not features-behind-a-paywall but the opposite: a transparent, git-native, terminal-first tool you point at any model. Aider also maintains a widely-cited polyglot LLM leaderboard that ranks frontier models on real code-editing tasks and — uniquely — publishes the dollar cost to run each one, making it a de facto reference for the price of AI code generation.

Pricing summary : How Aider’s pricing model works

Aider has no pricing in the conventional sense. The tool is free and open-source (Apache-2.0), so there is no list price, subscription, seat fee, or enterprise tier to capture. The single cost dimension a user faces is LLM API tokens — Aider connects to a model with the user’s own API key (bring-your-own-key), and the user pays the model provider’s published per-token rates directly. That spend scales entirely with usage: how many files are in the chat, how large the repository map is, and how chatty the session is.

• Software cost: $0 — free, open-source, no paywalled features.
• Token cost (the only real spend): pass-through LLM API usage, billed by the provider whose key you supply (OpenAI, Anthropic, Gemini, DeepSeek, xAI, Cohere, etc.).
• Zero-token-cost paths: local models via Ollama (run on your own hardware), OpenRouter’s free models (with daily usage limits), and Google’s Gemini 2.5 Pro Exp.

What makes this different: Aider externalizes its entire cost to a third party. There is no margin, no markup, and no Aider-side metering — the vendor never sees your money. It is the purest expression of token-based billing imaginable, except the meter belongs to the model provider rather than the tool. The pricing question for a prospective user is purely “which model will I point it at, and what does that model charge per token?” — which is why Aider also sits squarely in the pure-usage pricing pattern even though Aider itself charges nothing.

Pricing by product

Aider ships as a single product: the open-source CLI. There are no plan tiers to differentiate — the table below instead breaks down the one cost dimension (model tokens) by the kind of provider you connect.

Aider CLI (the tool)

Model access (your only real cost — BYO key)

Sales motions across products: pure self-serve / PLG — Aider is installed and configured by the user with no sales contact; there is no sales-led motion because there is nothing to sell. The only “purchase” is the model API key the user already holds.

Hidden costs : What Aider users actually pay

Aider’s “hidden cost” is the only cost: model API tokens. The software is genuinely $0, so the entire bill is whatever your chosen provider charges for the tokens Aider sends. The trap for new users is that this number swings by two orders of magnitude depending on which model you point Aider at — the same workflow can cost $1 or $150 with no change to the tool. The biggest token drivers are the repository map, the number of files added to the chat, large command-output pastes, and “architect/editor” two-model setups that send each request twice.

The clearest way to see the spread is Aider’s own polyglot benchmark, which runs the identical 225-problem suite on each model and publishes the dollar cost. That cost column is effectively a price sheet for “one fixed chunk of serious code editing”:

The same benchmark on DeepSeek ($1) versus o3-pro ($146) is a ~166x range — and both edit code competently. Now translate that to a month of real use.

Archetype A — a solo developer doing ~2 hours/day of AI pair programming on premium models. Assume each focused session is roughly the token weight of one benchmark run, ~20 working days/month:

Archetype B — the same developer, same workload, on DeepSeek (the cost-optimizer’s default).

So the identical Aider workflow runs from roughly $3/month to roughly $165/month purely on model choice — and the floor is $0 if you run a capable local model via Ollama or stay inside free tiers like OpenRouter free models or Gemini 2.5 Pro Exp. There is no platform fee absorbing any of this; the model choice is the pricing decision. This is exactly the kind of bill-shock and cost-unpredictability risk that catches teams who default to the most expensive model out of habit.

Want to estimate your own Aider bill? Use the Aider pricing calculator to model your monthly cost based on which model you connect, sessions per day, and average tokens per session.

Pricing evolution : how the cost of running a $0 tool collapsed

Aider has had exactly one “price” its entire life: $0. The earliest archived docs from mid-2023 already describe a free, open-source GPT-powered terminal tool with no subscription or seat fee, and that has never changed. So there is no price history to track — the meaningful evolution is the cost of the model tokens Aider depends on, which fell dramatically as competition and cheap reasoning models arrived. The cadence below tracks model-cost inflection points, not changes to Aider’s (nonexistent) price.

Cadence

Tracked range: 2023-07–2026-06. Aider’s own price was verified stable at $0 across the entire range (no price changes, no paid SKU ever added). “Price changes” above refer to Aider; the model-token cost moves are captured in Notes.

Notable changes

• 2023-07 — Earliest archived aider docs (Wayback snapshot of aider.chat/docs/faq.html) already show a free, open-source tool — confirming no paid tier ever existed.
• 2024-12-21 — Aider launches the polyglot benchmark, replacing its saturated Python-only test; its per-model cost column becomes a widely-cited reference for the price of frontier code editing.
• 2025-01-24 — Aider reports DeepSeek R1 + Sonnet set SOTA at ~14x lower cost, a concrete marker of the effective-cost crash for running aider.
• 2026-06 — Polyglot leaderboard shows a ~166x cost spread (DeepSeek ~$0.88 vs o3-pro ~$146.32) for identical work; Aider remains free and BYO-key.

The model-cost crash in detail

Because Aider externalizes its entire cost to the model layer, its “pricing story” is really the story of LLM token deflation. In 2023, running a serious AI coding session meant paying GPT-4’s premium API rates with no cheap alternative. The arrival of capable low-cost models — DeepSeek V3/R1, Gemini Flash, and OpenAI’s cheaper reasoning tiers — pulled the effective cost of the same Aider workflow down by one to two orders of magnitude. The polyglot leaderboard quantifies this precisely: the cheapest competent option (DeepSeek) now runs the full benchmark for under $1.50, while top-of-line OpenAI configs cost $30–$146. Aider didn’t change its price; the market changed it for them. This is the clearest case in the corpus of a tool whose value-to-cost ratio improved purely through upstream deflation — a dynamic explored in why token prices keep falling and what it means for AI pricing.

What’s unique : Aider’s distinctive pricing mechanics

1. The vendor never sees your money. Aider is the rare AI coding tool with no margin at all — no subscription, no markup on tokens, no usage tier. You pay the model provider directly with your own key, and Aider’s cut is exactly $0. Almost every competitor (Cursor, Copilot, hosted agents) buys tokens wholesale and resells them inside a seat fee or credit pool; Aider deliberately doesn’t, which makes its “pricing” a pass-through with no spread.

2. Cost is a model choice, not a plan choice. Because the meter belongs to the provider, the only lever a user has is which model to point Aider at — and that lever moves the bill across a ~166x range ($0.88 to $146 on the same benchmark). No other tool in the corpus puts the entire cost decision in the user’s hands like this. It turns “how much does this cost” into “which model do you trust and how much do you want to spend per token.”

3. A leaderboard that doubles as a price sheet. Aider’s polyglot leaderboard is unique in publishing a dollar-cost column next to each model’s accuracy. Most benchmarks rank only quality; Aider ranks quality and cost-to-complete, which makes it a widely-cited reference for the real price of frontier code editing — and a structural reason Aider stays cheap (it constantly surfaces the cheapest competent option).

4. Token-saving is the product’s cost control. Since Aider can’t discount tokens, its “cost controls” are engineering features: the compact repository map (context without dumping every file), --subtree-only and .aiderignore scoping, and confirmation prompts before pasting large outputs. These exist specifically to keep the provider bill down — cost management is built into the tool rather than into a billing dashboard.

5. A genuinely $0 floor. Local models via Ollama and free API tiers (OpenRouter free models, Gemini 2.5 Pro Exp) mean a determined user can run Aider indefinitely for $0 in API spend. Very few production-grade AI tools have a real, no-asterisk free path; Aider’s is structural, not a trial.

Strengths & weaknesses

Billing UX : Aider billing controls and transparency

Aider itself has no billing surface — no account, no invoice, no dashboard, because it never charges you. Cost controls are therefore expressed as token-saving features inside the tool, and the actual billing lives entirely with your chosen model provider.

• Token-cost controls in-tool — Aider’s repository map is built automatically to give the model context without adding every file, explicitly to limit token spend; the docs warn that adding extra files “will also increase your token costs.”
• Repo-map toggle — the repo map can be disabled by default for weaker models and force-enabled with --map-tokens 1024, directly tuning how many tokens each request consumes.
• Scoping controls — --subtree-only and a .aiderignore file (gitignore syntax) let you restrict Aider to part of a large/monorepo, reducing the context sent to the model and thus the token bill.
• Output confirmation prompts — Aider prompts before adding large blocks (e.g. “Add 6.9k tokens of command output to the chat?”) so you can decline costly context.
• Payment & invoicing — none from Aider. All payment, metering, and invoicing happen on the model provider’s side (OpenAI, Anthropic, etc.) via the API key you supply; local models via Ollama incur no API charges at all.

Strategic wins : Why Aider’s pricing decisions worked

1. Externalizing cost to the model layer

By charging nothing and making cost a pure pass-through, Aider sidestepped the hardest problem every hosted AI coding tool faces: how to price inference you don’t control while protecting margin as token prices swing. Aider has no margin to protect, so it simply rides upstream token deflation — every time models get cheaper, Aider gets cheaper for free, with zero pricing work. Competitors reselling tokens inside a seat fee have to constantly re-cut packaging to stay solvent; Aider never does. The mechanics of pure pass-through pricing are covered in usage-based pricing fundamentals.

2. Turning a benchmark into a moat

Aider’s polyglot leaderboard, with its dollar-cost column, made the project the reference for “what does frontier code editing actually cost.” That earned it durable distribution — its benchmark results repeatedly hit the Hacker News front page (the original Show HN reached 432 points), and “tops the aider leaderboard” became shorthand model vendors chase. A free tool can’t buy attention, so building the de facto measurement standard was a smart substitute for a marketing budget.

3. Making cost control a product feature, not a billing feature

Because Aider can’t discount tokens, it pushed cost management into the tool itself: the repo map, scoping flags, and large-output confirmation prompts all exist to keep the provider bill down. This is the right instinct even for vendors who do meter — the cheapest token is the one you never send, a lesson directly relevant to choosing the right usage metric and designing for efficient consumption rather than maximal billing.

4. A real $0 floor builds the funnel

Local models and free API tiers give Aider a no-asterisk free path, which is rare for production-grade tooling. That lowers the barrier to the very first try and lets cost-sensitive developers (students, OSS maintainers, hobbyists) adopt it without a payment relationship at all — the purest version of the product-led entry described in outcome-based and consumption pricing trends.

Areas to improve : Gaps in Aider’s pricing approach

1. Give users an in-tool spend estimate before each run

Aider’s biggest practical pain is unpredictability: the same task can cost $1 or $150, and the user only learns the difference after the provider bills them. A concrete fix is a pre-flight token/cost estimate (“this request will send ~12k tokens; on claude-opus-4 that’s ~$0.18”) shown before each model call, with a session running total. That turns the opaque pass-through into a visible meter and directly addresses the bill-shock and cost-unpredictability risk inherent to BYO-token tools.

2. Surface a “cheapest competent model” default

Aider already produces the data — the polyglot leaderboard ranks models by both accuracy and cost — but the tool doesn’t push new users toward the cost-optimal choice. A guided setup that recommends, say, DeepSeek or Gemini Flash for cost-sensitive users (with a one-line “≈100x cheaper than o3-pro for similar editing quality”) would prevent the most common overspend: defaulting to a premium model out of habit.

3. Add an optional spend cap / budget guard

Because all billing lives on the provider side, Aider has no way to stop a runaway session from racking up tokens. An optional client-side budget guard (“warn at $5/session, halt at $20”) would give users the safety rail that hosted competitors provide via plan caps — without Aider ever touching money. It would also make Aider far easier to sanction inside a team that needs predictable spend.

4. Document a team/enterprise BYO-key cost story

Aider has no procurement narrative, which limits org adoption even though the tool is excellent. Lightweight guidance on shared provider keys, per-developer cost attribution (Aider already commits with git attribution), and aggregating spend across a team would let engineering leaders adopt it deliberately rather than as shadow tooling.

Key takeaways

1. A $0 price can still be a pricing strategy. Aider’s decision to charge nothing and pass tokens through is a deliberate positioning choice, not an absence of one. It trades all revenue for maximum adoption and zero pricing-maintenance overhead — proof that “free, BYO-cost” is a viable model when your value is the workflow, not the inference.
2. Externalized cost rides upstream deflation for free. When your cost lives at a supplier you don’t mark up, every supplier price cut is an automatic improvement to your value proposition. Teams that resell a volatile input (tokens, compute) inside a fixed fee should study how much pricing work Aider avoids by simply not taking a spread.
3. Transparency about cost can be a distribution engine. Aider’s leaderboard cost column made it the reference for “what code AI costs,” earning recurring front-page attention no marketing budget could buy. Publishing honest cost data — even when it’s not your revenue — can build authority and trust.
4. The cheapest token is the one you never send. Aider’s repo map, scoping flags, and output prompts show that efficient context engineering is a pricing feature. Even vendors who meter should design to minimize wasted tokens rather than to maximize billable consumption.
5. Unpredictability is the cost of pure pass-through. The flip side of charging nothing is that the user inherits all the variance — a ~166x swing on model choice with no cap. Any team adopting a BYO-cost model needs to give users estimates, defaults, and guards, or they’ll feel the bill shock the vendor avoided.

UBP implications

1. Pure pass-through is the limit case of usage-based pricing. Aider shows what UBP looks like with the markup set to zero: the meter is real, the cost scales perfectly with consumption, but the vendor captures none of it. It’s a useful reference point for any pricing team deciding how much spread to take on a metered input — Aider takes 0%, hosted tools take 30–80%, and the trade-off is revenue versus adoption and trust.
2. When your input deflates, fixed-fee resale gets risky and pass-through gets safer. As token prices fall, vendors who bundled tokens into a flat seat fee must keep re-cutting packaging to avoid either overcharging or eroding margin. A BYO-token model like Aider’s is immune to that churn — a structural argument for pass-through (or transparent cost-plus) in fast-deflating markets.
3. Cost transparency is becoming a competitive feature, not just a courtesy. Aider’s published cost column raised the bar: buyers increasingly expect to see what a workflow actually costs across model choices. UBP vendors that expose real per-unit and per-workflow cost — rather than burying it in credits — will win trust as AI buyers get more cost-literate.

Sources

• Aider FAQ — open-source license, token-cost notes, repo-map controls (accessed 2026-06-08)
• Aider docs — Connecting to LLMs (free / local / paid model options) (accessed 2026-06-08)
• Aider LLM Leaderboards — polyglot benchmark with per-model dollar-cost column (accessed 2026-06-08)
• Aider blog — polyglot benchmark introduction (accessed 2026-06-08)
• Aider blog — R1 + Sonnet set SOTA at ~14x lower cost (accessed 2026-06-08)
• Aider official website (accessed 2026-06-08)

Bottom line

Aider is the corpus’s purest example of a $0 tool whose only cost is someone else’s meter: free, open-source, and BYO-token, with no subscription, markup, or margin. Its “pricing evolution” isn’t a price history at all — it’s the story of LLM token deflation pulling the effective cost of the same workflow from premium-only in 2023 to under $1.50 (or genuinely $0) in 2026. The lesson for everyone else: a transparent pass-through can be both a strategy and a moat, but it hands the user all the cost variance — so estimates, defaults, and guards are the missing product work.

Want to compare Aider against paid coding assistants like Cursor? Browse the full pricing blueprint.

Pricing timeline : Major events on a vertical axis

Each milestone below corresponds to a public pricing change, product launch, or material adjustment. Major events use a filled marker; minor adjustments use a faded one.

Facts captured — free OSS tool, BYO-key token cost confirmed

Jun 2026

Confirmed Aider remains free and open-source (Apache-2.0) with no subscription, seat, or enterprise pricing. The only user cost is pass-through LLM API tokens billed by the connected provider (BYO key). Polyglot leaderboard shows DeepSeek runs the benchmark for under $1.50 versus $10–$146 for frontier OpenAI models; $0-token paths exist via local Ollama models and free tiers.

DeepSeek R1 + Sonnet hits SOTA at ~14x lower cost

Jan 2025

Aider published that DeepSeek R1 as architect paired with Claude Sonnet as editor set a then-SOTA 64.0% on the polyglot benchmark at roughly 14x lower cost than the prior o1 result — a concrete marker of how the effective cost of running aider collapsed as cheap reasoning models arrived. Source: aider.chat/2025/01/24/r1-sonnet.html.

Polyglot leaderboard launches — cost becomes a first-class metric

Dec 2024

Aider replaced its saturated Python-only code-editing benchmark with the 225-problem polyglot benchmark (6 languages). Each model entry publishes a total-cost-in-dollars column, making the practical cost of running aider on each model transparent and widely cited. Source: aider.chat/2024/12/21/polyglot.html.

Free OSS from the start — no pricing page ever

Jul 2023

Earliest Wayback snapshot of aider's docs (2023-07) already describes a free, open-source GPT-powered terminal coding tool with no price, subscription, or seat fee. Aider has never had a pricing page; the only cost was always the user's own LLM API tokens. Source: web.archive.org snapshot of aider.chat/docs/faq.html.