The best GitHub Copilot alternatives, compared honestly
Copilot still has the best inline completion in the business, and GitHub does not meter it: completions and next edit suggestions stay unlimited on every paid plan. The meter is everywhere else. Since 2026-06-01 chat, the Copilot CLI, the cloud and coding agents and Copilot code review draw from a monthly pool of AI credits priced at $0.01 each, and on 2026-09-01 the transitional allowance that existing Business and Enterprise customers got for June, July and August drops from 3,000 and 7,000 credits per seat back to 1,900 and 3,900, with the seat price unchanged at $19 and $39 per granted seat per month. This page compares seven alternatives on the one axis that decides the bill: what the vendor counts, and whether you can pay the model provider instead.
The best GitHub Copilot alternative depends on which meter is hurting:
- You want to pay the model provider, not a middleman → Cline: Apache-2.0, $0 seat, inference at cost or your own key.
- You want completions and an agent in one editor → Cursor: unlimited Tab on Pro, Pro Plus and Ultra, Teams $40 per user per month billed monthly.
- You want the strongest terminal agent → Claude Code: Team from $20 per seat per month billed annually, though the size of its usage windows is not published.
- You want local models with no meter at all → JetBrains AI: your own Anthropic, OpenAI or OpenAI-compatible keys, plus unlimited use of local models through Ollama or LM Studio.
Narrower cases are in the table below: Zed if you want a GPL-licensed editor with published markup, OpenAI Codex if you live in ChatGPT limits, Amazon Q Developer if $19 a month per user and AWS billing settle it.
Why teams look elsewhere
What pushes teams off GitHub Copilot
Copilot did not get worse at the thing it is famous for. Completions are unlimited and unmetered on every paid plan, and GitHub does not mark up tokens. What changed is the arithmetic around them: one cutover on 2026-06-01, one allowance cut dated 2026-09-01, and a second meter on code review.
From 2026-09-01 the allowance shrinks and the seat price does not
GitHub's own billing doc: existing Business and Enterprise customers get a higher allowance “for the first three months of usage-based billing (June 1 – September 1, 2026)”, 3,000 and 7,000 credits per seat, after which “included usage returns to the standard amounts”, 1,900 and 3,900. Seats stay at $19 and $39 per granted seat per month. From 3,000 to 1,900 is 36.7% less; from 7,000 to 3,900 is 44.3% less. That percentage is our arithmetic on GitHub's two published figures, not a GitHub claim, and the boost covered existing customers only.
A Business seat includes exactly its own price in tokens
One AI credit is $0.01, so from 2026-09-01 a $19 Business seat includes $19 of model tokens and a $39 Enterprise seat includes $39. Individual plans get a flex allotment on top of base credits and therefore include more than they cost: Pro $10 per month includes $15, Pro+ $39 includes $70, Max $100 includes $200. There is no flex allotment for Business or Enterprise. The team plans are the only ones with zero headroom over their own price, and GitHub notes the flex allotment is variable by design and “may change over time”.
Code review is metered twice
GitHub states that “Code reviews have two cost components: AI credits for the model interaction (the review itself), and GitHub Actions minutes for the agentic capabilities (context gathering and tool use).” Published bands are $0.05 to $1 per review on Lite effort and $0.25 to $5 on Balanced, and both explicitly exclude the Actions minutes. GitHub states that “Model switching is not supported”, so a review cannot be moved to a cheaper model, and that consumption “generally increases with pull request size and repository custom instructions”. From 2026-09-01, four Balanced reviews at the top of GitHub's own range would exhaust an entire $19 seat, before anyone opens a chat window, which is our arithmetic on GitHub's published band. Actions minutes are charged on private repositories only; public repositories stay free, and self-hosted runners consume none.
Overage is on unless an admin turns it off
When an organization's pool runs dry, GitHub's default state is “Additional usage allowed: Usage continues at published per-credit rates”, and an administrator “must explicitly disable the AI credits paid usage policy” to prevent spend beyond the included credits. The reliable stop is a budget, because user-level budgets “always enforce a hard stop” and a $0 budget blocks a user immediately, with “no automatic fallback to lower-cost models when a budget is exhausted”. Credits never roll over; the pool resets at 00:00:00 UTC on the first of each month regardless of when licences were added.
Buying it is no longer fully self-serve
The plans doc states: “Starting April 22, 2026, new self-serve sign-ups for Copilot Business for organizations on GitHub Free and GitHub Team plans are temporarily paused.” Both Business and Enterprise are listed as contact-sales, no annual per-seat price is published for either, and Copilot “is not currently available for GitHub Enterprise Server”, which rules it out for air-gapped estates. Individual sign-ups were paused and then reopened per a changelog dated 2026-06-17.
Developers report burning the pool in hours
User reports, not vendor facts. GitHub's own community discussion #198015, “GitHub Copilot AI Credits Are Unfair, Expensive, and Killing Real Development Workflows”, dated 2026-06-04, carries about 240 upvotes, 89 comments and 71 replies. Representative claims from participants: “I used 100% of copilot pro+ in just 2 days of normal work” and “I burned through it in a matter of HOURS”. Treat these as individual experiences rather than measurements; they are useful only as a signal of which workflows hit the pool fastest.
The shortlist
7 GitHub Copilot alternatives worth evaluating
Copilot is really two products on one invoice: an unmetered completion engine and a metered agent. Almost nothing on this list replaces both. The ranking optimises for a meter you can predict and an exit you can take, so a tool that lets you pay the model provider directly ranks above one that resells inference. Each pick lists one honest strength and one real weakness.
Apache-2.0, verified in the LICENSE header, and the busiest repo in this set: about 66,500 stars with a push and a release on 2026-08-21. The plan table has two rows, “Open Source” at $0 and “Enterprise” at a price Cline does not publish, because the product is free and you buy inference: “Purchase and Model Inference at Cost or BYOK”. Cline's blog claims you “pay exactly what the model provider charges, nothing more”, which is a vendor claim on a blog rather than a rate on the pricing page. Weakness: no inline completion at all, so it replaces the agent half of Copilot and nothing else.
Pro $20 per month or $16 per month billed annually; Pro Plus $60 or $48; Ultra $200 or $160; Teams Standard $40 per user per month, $32 billed annually. Pro, Pro Plus and Ultra include unlimited Tab completions, the one thing most Copilot refugees actually miss. The meter is two-layered: two separate monthly usage pools (Cursor's own models and third-party models), each resetting with the billing cycle and with no transfer between them published, and a $0.25 per million token Cursor Token Rate on Teams and Enterprise that applies on top of model API pricing. Weakness: the Teams allowance is not published in dollars, and its BYOK cannot drive Tab, is excluded from Cursor's Zero Data Retention policy and still pays that $0.25 per million on Teams.
Pro is $17 per month billed annually ($200 up front) or $20 billed monthly; Max is $100 or $200 per month; Team Standard is $20 per seat per month billed annually, $25 monthly. Billing is a seat plus included-usage windows, “a per-seat allowance that resets on a rolling five-hour window and a weekly window”, shared with Claude chat and Cowork. Overage bills at standard API rates, and Anthropic's own anchor is “around $13 per developer per active day and $150-250 per developer per month”. BYOK is real through Bedrock, Google Cloud and Microsoft Foundry. Weakness: the absolute window size is published nowhere, only as “5x” and “20x” multipliers of an unstated base, and its LICENSE.md is a copyright notice, not an open-source licence.
Go $8 per month, Plus $20, Pro from $100, Business $20 per user per month billed annually for two or more users, or $25 monthly. The CLI client is genuinely Apache-2.0 while the service is proprietary, and BYOK is offered: “Pay only for the tokens Codex uses, based on API pricing.” Limits are message ranges per rolling five-hour window, and OpenAI states that “The usage limits for local messages and cloud chats share a five-hour window”, for example 10 to 100 Sol messages on Plus. Weakness: those ranges are ten times wide, weekly caps are only referenced as “additional weekly limits may apply” with no numbers, Free and Go limits are unpublished, and no dollar value for a ChatGPT credit is published at all (the $0.04 figure often quoted is derived, not official).
The editor and its collab server are GPL-3.0-or-later, with Apache-2.0 for the gpui framework, all OSI-approved, and the repo pushed on 2026-08-21 with about 89,000 stars. Personal is $0 forever with “unlimited use with your API keys or external agents”; Pro is $10 per month and includes $5 of tokens; Business is $30 per seat per month. Zed is also the rare vendor that publishes its markup: hosted usage bills at “API list price +10%”. Weakness: it is a whole new editor rather than an extension, Business bundles no token credit at all, and Zed is the only tool here that meters completions, capping the free plan at 2,000 accepted edit predictions.
Completions are unlimited on every tier including Free, and the escape is the most complete in this set: your own Anthropic, OpenAI or OpenAI-compatible keys, plus “unlimited use of your own local models, connected via Ollama, LM Studio, or another OpenAI-compatible server”. Credits are the mirror image of GitHub's, one credit being $1 of model spend rather than $0.01, and on JetBrains' organization prices AI Free gets 3 credits, AI Pro at $20 per user per month gets 20, AI Ultimate at $60 per user per month gets 70, and AI Enterprise, also $60 per user per month in the same table, is sold on a yearly cycle only. The quota “reset occurs every 30 days from the date when you first used your JetBrains AI license”, not on a calendar month. An All Products Pack or dotUltimate subscription already includes AI Pro. Weakness: JetBrains will not disclose the AI Enterprise credit count, saying only that it is “on par with AI Ultimate quota, or higher”, and two of its doc sets disagree on the free tier, the AI Assistant docs calling it 3 credits per 30 days while the older JetBrains AI service doc still calls it “a one-time, non-renewable limited quota”.
Pro is $19 per month per user, billed monthly, the same headline as Copilot Business, and the free tier is concrete: 50 agentic requests per month plus 1,000 lines per month of Java transformation. The unit is refreshingly plain, “any Q&A chat or agentic coding interaction with Q Developer through either the IDE or Command Line Interface”. Weakness: Pro's allowance is described only as “Included (with limits)” with no number and no per-request overage rate, the only published per-unit price is $0.003 per line of code submitted for Java-upgrade transformation once the account’s pooled allocation of 4,000 lines per subscription per month is spent, there is no documented BYOK path, and AWS says the monthly fee starts when a user first performs activity with Q Developer and that cancelling before month end is still charged the full month's fee.
Deliberately not shortlisted, with reasons: Devin Desktop, the product formerly sold as Windsurf, because windsurf.com/pricing now 308-redirects to devin.ai/pricing and no plan's quota quantity is published anywhere, not even for Free, while its own quota doc grandfathers existing Pro subscribers at $15 per month and existing Teams seats at $30 per seat per month indefinitely, against new list prices of $20 per month and $40 per full seat per month, which is a 33% increase by our arithmetic. Tabnine, acquired by Tricentis, because its free tier is gone and its two remaining plans, at $39 and $59 per user per month, are sold on annual subscription only. Aider, Apache-2.0 and 48,371 stars, because it has shipped no release since 2025-08-09. Roomote, the successor to Roo Code, because Roo Code is archived, its last commit and final release both dated 2026-05-15 and the successor is source-available under FCL-1.0-ALv2, not open source. Continue, because it was acquired by Cursor: its own page title reads “Continue (acquired by Cursor)”, its /pricing returns 404 and its hosted hub is gone, so listing it beside Cursor as an independent alternative would be misleading. Three of six open-source candidates changed hands or stalled inside twelve months, which is a reason to weigh survival as heavily as features.
Side by side
GitHub Copilot alternatives compared
Prices as of August 2026, from each vendor's own pricing page or docs, with the billing cycle labelled on every figure. The column that decides most migrations is Bills by, normalized: two vendors have independently made a credit a unit of currency 100 times apart, GitHub at $0.01 and JetBrains at $1.00, while the rest meter time windows, messages or requests. A cell reading not published means the vendor publishes no number, which is a finding rather than a gap. GitHub Copilot's own row is the baseline.
| Tool | Bills by (normalized) | Seat price, cycle labelled | Free tier's real limit | BYOK | Licence | Best for |
|---|---|---|---|---|---|---|
| GitHub Copilot | Dollars of tokens: 1 AI credit = $0.01, and Business and Enterprise include 100% of the seat price from 2026-09-01 | Pro $10/mo · Pro+ $39/mo · Max $100/mo · Business $19 per granted seat/mo · Enterprise $39 per granted seat/mo. No annual per-seat price published | 2,000 completions/mo; credit allowance not published; code review excluded | ✓ documented: local BYOK in VS Code, JetBrains, Xcode, the CLI, the app and the SDK; enterprise keys in public preview | ✗ proprietary SaaS | Teams whose usage is mostly completions, and who want IP indemnity |
| Cline | Inference at cost, or your own key. No seat meter exists | Open Source $0 · Enterprise not published | The whole tool, unlimited; you pay only inference | ✓ fully | ✓ Apache-2.0 (OSI) | Teams that want the meter out of the loop entirely |
| Cursor | Two separate usage pools, first-party and third-party, each resetting with the billing cycle and with no transfer between them published, plus $0.25/1M Cursor Token Rate on Teams and Enterprise | Pro $20/mo or $16/mo annual · Pro Plus $60/mo or $48/mo annual · Ultra $200/mo or $160/mo annual · Teams Standard $40/user/mo or $32/user/mo annual · Teams Premium $120/user/mo or $96/user/mo annual | “Limited Agent requests”: not published | ✗ partial: no Tab, outside Cursor's Zero Data Retention policy, Teams still pay $0.25/1M | ✗ proprietary | Keeping unlimited Tab completion after leaving Copilot |
| Claude Code | Seat plus rolling five-hour and weekly usage windows, shared with Claude chat and Cowork | Pro $17/mo annual or $20/mo monthly · Max $100 or $200/mo · Team $20/seat/mo annual, $25 monthly | Free $0; absolute limit not published, only “5x” and “20x” | ✓ Bedrock, Google Cloud, Microsoft Foundry | ✗ not open source: Anthropic Commercial Terms | Long agentic runs in the terminal |
| OpenAI Codex | Messages per rolling five-hour window, published as ten-times-wide ranges | Go $8/mo · Plus $20/mo · Pro from $100/mo · Business $20/user/mo annual, $25 monthly | Free $0; limit not published; weekly caps not published | ✓ API-priced tokens | Client Apache-2.0; service proprietary | Teams already standardised on ChatGPT plans |
| Zed | Dollars of tokens at “API list price +10%”, plus a separate accepted-completions cap | Personal $0 · Pro $10/month · Business $30/seat/month. No annual cycle published | 2,000 accepted edit predictions, plus unlimited use with your own keys | ✓ fully, on the free plan | ✓ GPL-3.0-or-later (+ Apache-2.0 for gpui) | Buyers who want an OSI licence and a published markup |
| JetBrains AI | Credits where 1 credit = $1.00 of model spend, reset on a rolling 30 days | Organization prices from the docs table: Free $0 · AI Pro $20/user/mo · AI Ultimate $60/user/mo · AI Enterprise $60/user/mo, billed yearly only | 3 credits per 30 days, no top-up; JetBrains' own docs also call it a one-time quota | ✓ own Anthropic, OpenAI and OpenAI-compatible keys, plus unlimited local Ollama and LM Studio models | ✗ proprietary | JetBrains shops, and anyone running local models |
| Amazon Q Developer | Agentic requests: any chat or agentic coding interaction, IDE or CLI | Pro $19/month per user, billed monthly | 50 agentic requests/month, plus 1,000 lines/month of Java transformation | ✗ no documented path | ✗ service proprietary; CLI repo dual-licensed MIT or Apache-2.0 | AWS-native teams billing through their AWS account |
Numbers that do not fit in cells: Copilot pools credits at the billing entity, so “an enterprise with 100 Copilot Business users gets a shared pool of 190,000 AI credits rather than 100 individual buckets”, and individual plans have no pooling. Copilot's third-party agents, which GitHub names as Anthropic Claude and OpenAI Codex, run inside Copilot on Pro, Pro+, Max, Business and Enterprise and consume AI credits and GitHub Actions minutes, so a competitor's agent can run on GitHub's meter. Comparing dollars per million tokens across vendors is not apples to apples: Anthropic's token-counting doc says that on Claude 4.7 and later models “the same input text produces approximately 30 percent more tokens than on earlier models”, and that usage and billing reflect those counts. Two GitHub model promotions expire in the same window as the credit boost, GPT-5.6 Sol at 50% off through 2026-09-03 and Gemini 3.6 and 3.7 Flash at a promotional $0.75 per million input tokens and $3.75 per million output through 2026-12-31. Pricing and free tiers change often; check each vendor for current terms. Compiled August 2026.
Official pages: GitHub Copilot billing · GitHub model rates · Cline · Cursor · Claude · ChatGPT and Codex · Zed · JetBrains AI · Amazon Q Developer
The ten-seat calculator
What ten seats actually buy
The only unit these vendors share is dollars of model tokens: three of them meter in it natively and every overage falls back to it. So the honest comparison is what ten seats cost per month and how much model spend those seats include. Where a vendor publishes no allowance, the cell says so. A calculator that hides a guess is worse than no calculator, so nothing below is estimated: every figure is either published by the vendor, arithmetic on published figures, or marked not published.
| Plan, 10 people | Seats per month, monthly cycle | Seats per month, annual cycle | Included model spend for 10 | Where the numbers come from |
|---|---|---|---|---|
| Copilot Business | $190 | not published | $190 of tokens from 2026-09-01, our arithmetic on 1,900 credits × 10 seats; $300 under the promotional allowance that ran from 2026-06-01 to 2026-09-01, existing customers only | Published: seat price, credit count, $0.01 per credit |
| Copilot Enterprise | $390 | not published | $390 from 2026-09-01, our arithmetic on 3,900 × 10 seats; $700 under the same promotional allowance, which ran to 2026-09-01 | Published, same three sources |
| Cursor Teams Standard | $400 | $320 | Not published in dollars; Premium is documented only as adding “5x the Standard limits on Agent” | Seat published; annual price read from the rendered pricing page on 2026-08-21; allowance a gap |
| Cursor Teams Premium | $1,200 | $960 | Not published: documented only as “5x the Standard limits on Agent” | Multiplier of an unpublished base |
| Claude Code Team Standard | $250 | $200 | Usage windows; size not published | Seat published both cycles; window size a gap |
| Claude Code Team Premium | $1,250 | $1,000 | As above | Seat published both cycles; the enterprise page states $100 per person per month with a two-member minimum |
| Codex Business | $250 | $200 | Plus-level ranges per five-hour window per user, for example 10 to 100 Sol messages; weekly caps not published | Ranges published; dollars per credit derived, never published |
| Zed Pro | $100 | not published | $50 ($5 × 10), then API list price +10% | Published, including the markup |
| Zed Business | $300 | not published | $0: Business bundles no token credit | Published |
| JetBrains AI Pro | $200 | $2,000 per year, which is $166.67 per month | $200: 20 credits × 10 seats × $1, our arithmetic | Organization price and cycle from JetBrains' buy page; credit counts from its help docs |
| JetBrains AI Ultimate | $600 | $6,000 per year, which is $500 per month | $700: 70 × 10, our arithmetic, and the only plan here whose included spend exceeds its seat cost | As above |
| Amazon Q Developer Pro | $190 | monthly only | Not published: “Included (with limits)”, no request count, no overage rate | Seat published; allowance a gap |
| Devin Desktop Teams | $400: ten full seats at Devin's published $40 per seat per month, our arithmetic, above its published $80 monthly minimum | not published | Quota not published for any plan, including Free | Seat arithmetic published; quota quantity a gap. Enterprise bills in ACUs at a rate set per contract, also not published |
| Cline | $0 | $0 | None bundled: you buy inference at cost or bring your own key | Published; the 0% markup figure appears on Cline's blog, not its pricing page |
| Roomote Cloud | $49 flat for the whole band up to 10 users | not published | $0: “Us: Hosting, networking, sandboxes, upgrades / You: Inference key” | Published at roomote.dev, whose /pricing URL 404s |
| Aider | $0 | $0 | $0: no meter exists | Published by absence: no pricing page, no billing link |
Cost of one agent run, where anyone publishes it
| Rate | Value | Status |
|---|---|---|
| Copilot code review, Lite effort | $0.05 to $1 of AI credits per review | Published by GitHub |
| Copilot code review, Balanced effort | $0.25 to $5 per review, so from 2026-09-01 four reviews at the top of the range spend a whole $19 seat | Published range; the multiplication is ours |
| The second meter on those reviews | GitHub Actions minutes, excluded from the estimates above, private repositories only. Default Linux 2-core runner: $0.006 per minute, rounded up per job | Published |
| Copilot review model | “Model switching is not supported, as changing the model is likely to compromise reliability, user experience, and the quality of review comments”, so you cannot move a review to a cheaper model or pre-compute its token cost | Published as a restriction |
| Codex, per message | “GPT-5.6 usage averages 5-40 credits per message”, which is $0.20 to $1.60 at $0.04 per credit | Range published; the dollar conversion is derived, never published by OpenAI |
| Cursor Auto Cost mode | $1.25 per 1M input, $1.25 per 1M cache write, $0.25 per 1M cache read, $6.00 per 1M output | Published |
| Cursor Token Rate | $0.25 per 1M on input, output and cached tokens, Teams and Enterprise, including your own keys | Published |
| Devin quota per run | Not published for any plan | Gap |
| Claude Code spend anchor | “Around $13 per developer per active day and $150-250 per developer per month” | Vendor estimate, not a rate |
Cells this calculator refuses to fill, because the vendor publishes no number: Devin's dollars per ACU and ACUs included per plan; Devin and Windsurf quota quantity on every plan including Free; Cursor's Teams included usage in dollars; Cursor's Hobby agent-request count; Claude Code's absolute usage-window size on every plan; Codex weekly limits; Codex Free and Go limits; Amazon Q Pro's agentic-request allowance and per-request overage; Copilot Free's credit allowance; Copilot Business and Enterprise annual per-seat price; Cline's Enterprise price and signup-credit amount; Tabnine's self-hosted, VPC and air-gapped price and its monthly cycle; Roomote's 100-plus-user bands and annual cycle; and any price at all for Continue. Fourteen unpublished cells in a category that sells itself on measurable productivity is itself the finding.
A fair call
When GitHub Copilot is still the right choice
Migrating costs more than the difference in list price: a new editor or extension per developer, new keys and budgets to administer, and, if you go BYOK, an inference bill that lands on a card instead of an existing invoice. Copilot also has one honest advantage that most of its critics skip past. GitHub does not mark up tokens; its published rates match the list prices of the two providers we checked, Anthropic and OpenAI, model by model. The squeeze is a shrinking allowance, not a margin grab, and a shrinking allowance is something you can plan around.
GitHub Copilot is still right if…
- Your team's usage really is mostly completions. They are unlimited and unmetered on every paid plan, and no other vendor here states that as plainly; Zed actively caps them on its free tier.
- You want tokens at provider list price with no handling fee. Zed adds 10% and Tabnine 5%; GitHub adds nothing, and gives a further 10% discount on model costs while using auto model selection.
- Your reviews are light or your repositories are public. The double meter on code review only charges Actions minutes on private repos, and self-hosted runners consume none.
- You need pooling. Credits pool at the billing entity, so 100 Business seats share 190,000 credits instead of 100 separate buckets, and heavy users absorb light ones. Individual plans get no pooling at all.
- You need IP indemnity, which GitHub names as one of the differences between its organization and individual offerings.
- You want the competitor's agent without a second contract: the agents GitHub names as Anthropic Claude and OpenAI Codex run inside Copilot on Pro, Pro+, Max, Business and Enterprise, on GitHub's meter.
- Vendor survival matters more than list price. Roo Code was archived, Continue was acquired, Tabnine was acquired and lost its free tier, and Aider has not shipped a release since 2025-08-09.
Look elsewhere if…
- Your bill is driven by how many pull requests you review rather than how much code you write: Cline or Zed put you back on provider rates.
- You run on GitHub Enterprise Server, where GitHub states Copilot “is not currently available”. For air-gapped work GitHub does now document local BYOK as suitable, but JetBrains AI with Ollama or LM Studio is the option that keeps inference local on every tier including Free.
- You need to know the number before you buy. JetBrains publishes credits per tier and Zed publishes both its included dollars and its markup; Copilot Business and Enterprise publish no annual per-seat price and cannot always be bought self-serve.
- You want a hard ceiling by default rather than by policy. Copilot's organization overage is enabled unless an admin disables it, and only a user-level budget hard-stops.
- You are on an individual plan and object to training on your interactions. GitHub states that it does not use Copilot Business or Copilot Enterprise customer data to train AI models, and that for individual subscribers, meaning Copilot Free, Pro, Pro+ and Max, it may use interaction data including prompts, suggestions and code snippets generated during Copilot sessions to train and improve its models. Individual subscribers can opt out.
- You want an OSI licence you can audit: Cline is Apache-2.0 and Zed is GPL-3.0-or-later.
Common questions
Common questions about GitHub Copilot alternatives
Does GitHub Copilot still have unlimited usage?
Code completions are still unlimited on every paid Copilot plan. Everything else is not. Since 2026-06-01, chat, the Copilot CLI, the cloud agent, Spaces, Spark and third-party coding agents draw from a monthly pool of AI credits, where one credit equals one US cent of model tokens. GitHub's documentation is explicit: “Code completions and next edit suggestions are not billed in AI credits. They remain unlimited for all paid plans.” Completions being unmetered is why the change is easy to miss. You can write code all month without touching the meter, then exhaust it reviewing other people's pull requests.
What changes for Copilot Business and Enterprise on 2026-09-01?
The promotional credit boost for existing Copilot Business and Copilot Enterprise customers ends. GitHub's documentation says those customers “receive a higher amount of included AI credits for the first three months of usage-based billing (June 1 – September 1, 2026)”, namely 3,000 credits for Business and 7,000 for Enterprise, and that after the promotional period included usage returns to the standard amounts, which are 1,900 and 3,900. Seat prices do not change: $19 and $39 per granted seat per month. On GitHub's two published figures that is 36.7% and 44.3% less included usage at the same price. The boost covered existing customers only.
What does a Copilot code review actually cost?
GitHub publishes a range: $0.05 to $1 worth of AI credits with “Lite” effort and $0.25 to $5 with “Balanced” effort. GitHub states that “Code reviews have two cost components: AI credits for the model interaction (the review itself), and GitHub Actions minutes for the agentic capabilities (context gathering and tool use).” Those minutes apply to private repositories; public repositories stay free. The published dollar estimates explicitly exclude the Actions minutes, GitHub states that for Copilot code review “Model switching is not supported”, so you cannot move a review to a cheaper model, and that consumption “generally increases with pull request size and repository custom instructions”. From 2026-09-01, four reviews at the top of GitHub's own Balanced range would consume a $19 Business seat's entire monthly allowance; that multiplication is our arithmetic, not a GitHub claim.
What happens when the AI credits run out?
For organizations, additional usage continues at $0.01 per credit and it is enabled by default: an administrator has to explicitly disable the AI credits paid usage policy to stop the spend. Budgets are the hard stop. GitHub says user-level budgets “always enforce a hard stop”, a $0 budget blocks a user immediately, and “There is no automatic fallback to lower-cost models when a budget is exhausted.” Credits never roll over, and the pool resets to the full monthly amount at 00:00:00 UTC on the first day of each calendar month. Individuals on Pro or Pro+ must set their own budget, and GitHub says any additional usage already consumed has to be paid off before use continues.
Is GitHub marking up the model tokens?
No, and that matters when you pick an alternative. GitHub's published Copilot token rates match the list prices of the two providers we checked, Anthropic and OpenAI, model by model: Claude Opus 5 at $5 per million input and $25 per million output tokens, Claude Sonnet 5 at $2 and $10, GPT-5.6 Terra at $2 and $12 on its default tier of up to 272K input tokens. Copilot's cost pressure is the size of the included allowance, not a margin on inference. Some alternatives do publish a markup: Zed bills “API list price +10%” and Tabnine bills “based on actual LLM provider prices + 5% handling fee”, while Cline sells inference at cost.
Which alternatives let me pay the model provider directly?
Four do it cleanly, and GitHub Copilot is one of them. GitHub documents bring-your-own-key for Copilot: local BYOK handles your own API keys client-side in VS Code, JetBrains, Xcode, the Copilot CLI, the Copilot app and the SDK, described as suitable for air-gapped environments or users without a Copilot subscription, with enterprise-managed keys in public preview. Zed's free Personal plan allows unlimited use with your own API keys or external agents. JetBrains IDEs accept your own Anthropic, OpenAI and other OpenAI-compatible keys, and JetBrains offers unlimited use of your own local models connected through Ollama, LM Studio or another OpenAI-compatible server. Cline is bring-your-own-key by default, or sells inference at cost. Aider and Roomote have no inference meter at all. Cursor's own-key support is partial: custom keys do not work with Tab completion, Cursor states that its Zero Data Retention policy “does not apply when you use your own API keys”, and on Teams and Enterprise plans its $0.25 per million token rate still applies to BYOK usage. Amazon Q Developer publishes no bring-your-own-key path.
Is there a genuinely open-source GitHub Copilot alternative?
Fewer than the marketing suggests, and the licence distinctions are real. Genuinely OSI-licensed: Cline and Aider under Apache-2.0, the Codex CLI client under Apache-2.0 although the service behind it is proprietary, Zed under GPL-3.0-or-later with Apache-2.0 for its gpui framework, and Continue under Apache-2.0. Source-available but not open source: Roomote, the successor to the archived Roo Code, under FCL-1.0-ALv2, which bars “Competing Use” and converts each release to Apache-2.0 after two years. Not open source at all: Copilot, Cursor, Claude Code, whose LICENSE.md subjects use to Anthropic's Commercial Terms, Tabnine, whose MIT-labelled repository README says “There are no source files here because the backend is closed source.”, JetBrains AI, Amazon Q Developer and Devin.