Copilot Pricing: $10 Pro Buys 1,500 AI Credits [2026]

GitHub's Copilot Pro plan at $10 a month includes 1,500 AI Credits priced at $0.01 each, and VS Code now runs GitHub's own coding agent alongside Anthropic's Claude and OpenAI's Codex in a single Agents window, according to VS Code documentation and GitHub's changelog. The credit system dates to June 1, 2026, when third-party trackers say GitHub replaced premium-request quotas with credits, and the October stack adds Claude Opus 5.5, Sonnet 5.5 and Haiku 5.5 in the model picker, GPT-6.1 Sol from the September 28 weekly release, and generally available local sandboxing in the October 5 release at no extra charge. Agent automations and Agent Merge remain in preview, and the reported VS Code 1.141 version number appears in only one source.

GitHub Copilot in VS Code now runs three coding agents side by side: GitHub’s own, Anthropic’s Claude and OpenAI’s Codex, according to the VS Code documentation https://code.visualstudio.com/docs/supporting/faq . Every one of those sessions draws on a single meter. GitHub AI Credits cost $0.01 each, and the Copilot Pro plan at $10 a month ships with 1,500 of them. That pairing is why GitHub Copilot pricing now deserves a second look from anyone who budgets for developer tools. The pricing change itself dates to June 1, 2026, when third-party trackers say GitHub swapped premium-request quotas for credits. What is new in October is the stack on top of it. Claude Opus 5.5, Sonnet 5.5 and Haiku 5.5 sit in the model picker. GPT-6.1 Sol arrived with the September 28 weekly release. Local sandboxing went generally available in the October 5 release at no extra charge. Meanwhile the standalone tools keep moving, as the recent run of Claude Code and Codex CLI updates https://shattered.io/claude-code-codex-cli-opencode-update-4-days-2026/ showed. For a team lead, the question is no longer which chatbot writes better code. It is which agent to standardize on, how many credits a normal week burns, and who pays when an autonomous session runs overnight. The numbers below come from GitHub’s changelog, Anthropic’s and OpenAI’s own pages, and a small set of third-party trackers that are flagged as such. Don't miss new tech stories on Google Add Shattered once in the Google app and our stories show up in your news suggestions. Add now https://www.google.com/preferences/source?hl=en&q=https%3A%2F%2Fshattered.io%2F Three Agents, One Window: What Changed in VS Code VS Code now treats agents as first-class citizens. The Agents window hosts Copilot, Claude and Codex sessions in the same list, and each can open a pull request without leaving the editor. GitHub documented the pull-request flow on October 1 in its roundup of the September releases, which covered VS Code versions 1.136 through 1.140. A developer reviews the title, edits the description, picks draft or merge options, and submits. Codex got a continuity feature in the same batch. A conversation started in the ChatGPT app can continue inside VS Code, and nobody has to copy context or files by hand. Isolated Git worktrees for Copilot, Claude and Codex sessions arrived earlier, on July 30, along with a Claude change that lets one agent session hold several related chats, each with its own history, title and model. Two September features remain in preview. Agent automations schedule recurring work hourly, daily or weekly, or run it on demand. Agent Merge lets an agent handle review feedback, failed checks, merge conflicts and workflow reruns so a pull request is ready to land. Both sit behind preview labels, so teams should treat them as experiments rather than production dependencies. On October 5, the Agents window added side-by-side sessions, and chat gained worktree cleanup controls. GitHub’s release notes for that week point to VS Code 1.141, though that version number appears in only one source, so treat it as reported rather than confirmed. Agents also run inside Dev Containers over SSH, tunnels and WSL hosts as of September 25, with a gradual rollout. A local Dev Container path needs Docker and a supported configuration, per the September 18 release. How AI Credits Meter Copilot Pricing GitHub’s own pricing page describes AI Credits as the way developers pay for AI usage in Copilot. Every plan includes a monthly allowance, and one credit equals $0.01. A $10 overage budget therefore buys 1,000 credits. Copilot settings show usage and the reset date, with alerts at 75%, 90% and 100% of any budget a user configures. One independent breakdown of Copilot AI Credits https://www.danilchenko.dev/posts/copilot-ai-credits/ reports that charges follow model input, output and cached tokens. That matters because a long agent session reads far more context than a short chat. A cached read costs less than a fresh one, so session design affects the bill as much as model choice does. What counts against your credits Chat, agents, Copilot CLI and Spaces consume credits. Opt-in secret-classifier checks for push protection and the security-review command also use them, starting after a team opts in to the public preview. GitHub announced that change on October 7. Code review draws on credits too, and since June 1 the code review workflow also eats GitHub Actions minutes, according to GitHub’s pricing notes. What stays free Code completions and next-edit suggestions do not consume credits on paid plans. That split shapes behavior. Developers who mostly accept inline suggestions will barely touch their allowance, while those who hand whole tasks to agents will watch the counter move. Local sandboxing is also free, and GitHub states no additional charge for it. Copilot Pricing by Plan: Credits Per Dollar GitHub lists six plan names for individuals and organizations: Free, Pro, Pro+, Max, Business and Enterprise, plus a free Student tier. The monthly credit totals combine a base allocation with a flexible one. The table below uses the figures GitHub’s documentation and pricing page show today. | Plan | Monthly price | Monthly AI Credits | Credit value | Notes | |---|---|---|---|---| | Copilot Free | $0 | Allowance included | Not stated | Limited agents, automatic model selection only | | Copilot Student | $0 | Allowance included | Not stated | Third-party agents excluded | | Copilot Pro | $10 | 1,500 1,000 base + 500 flexible | $15 | Entry paid tier | | Copilot Pro+ | $39 | 7,000 3,900 base + 3,100 flexible | $70 | Gets the newest premium models first | | Copilot Max | $100 | 20,000 10,000 base + 10,000 flexible | $200 | Built for sustained agent workflows | | Copilot Business | $19 per seat | Shared organization pool | Pooled | Overage at $0.01 per credit | | Copilot Enterprise | $39 per seat | Shared organization pool | Pooled | Overage at $0.01 per credit | The arithmetic shows a clear ladder. Pro returns $15 of credits for $10, a 1.5x ratio. Pro+ returns $70 for $39, roughly 1.8x. Max returns $200 for $100, a clean 2x. Heavy agent users therefore pay less per unit of model usage as they climb, which is the opposite of how most per-seat software prices. One caution applies. The Free and Student rows state an allowance without a published figure on the pages GitHub serves today, so those cells say so rather than guess. Third-party snippets sometimes quote different numbers, and GitHub’s own documentation should settle any dispute. Which Models Copilot Offers Now The model list turned over fast in the past seven weeks. Claude Fable 5.1 landed on September 4, Claude Opus 5.5 on September 25, and GPT-6.1 Sol with the September 28 release. Claude Sonnet 5.5 followed on October 2, Haiku 5.5 appeared in the October 9 changelog, and Google’s Gemini 4 Argon went generally available in October for complex coding, agentic and reasoning work, per GitHub’s October changelog https://github.blog/changelog/month/10-2026/page/2/ . | Model | Vendor | Reached Copilot | Paid plans listed | |---|---|---|---| | Claude Fable 5.1 | Anthropic | September 4 | Pro+, Max, Business, Enterprise | | Claude Opus 5.5 | Anthropic | September 25 | Pro+, Max, Business, Enterprise | | GPT-6.1 Sol | OpenAI | September 28 release | Pro+, Max, Business, Enterprise | | Claude Sonnet 5.5 | Anthropic | October 2 | Pro, Pro+, Max, Business, Enterprise | | Claude Haiku 5.5 | Anthropic | October 9 changelog | Pro, Pro+, Max, Business, Enterprise | | Gemini 4 Argon | | October generally available | Listed as generally available | The pattern is plain. Premium Claude and GPT models reach Pro+ and above first, and the small, cheap models reach Pro on day one. Developers on the $10 plan get Sonnet 5.5 and Haiku 5.5, but not Opus 5.5, according to GitHub’s changelog entries for each release. GitHub also opened a research preview called HydraFusion on September 28 in VS Code, the Copilot app and Copilot CLI. It lets Copilot select and coordinate models for a task, so a developer no longer picks one model per prompt. It is a preview, not a generally released feature, and GitHub has not published how it prices a coordinated multi-model run. Retired models Supply shrank as fast as it grew. According to GitHub’s changelog roundup, the company retired six models on October 19 https://shattered.io/github-copilot-retires-6-models-gpt-5-6-2026/ : Gemini 3.7 Flash, GPT-5.5, GPT-5.4, GPT-5.4 mini, GPT-5 mini and Grok 4.5. GitHub listed a replacement for each. An earlier October 2 deprecation round pulled several other models across chat, inline edits, ask mode, agent mode and completions. Teams with pinned model names in custom agents or prompts should audit them now. Claude Code Inside and Outside the Copilot Meter Anthropic’s current model family is Opus 5.5, Sonnet 5.5 and Haiku 5.5. The company says Opus 5.5 leads in agentic coding and knowledge work and costs 40% less to run than Opus 5 on typical workloads. Sonnet 5.5 runs more than 30% faster than Sonnet 5 and costs up to 30% less for most work. Haiku 5.5 is described as the fastest and most capable small model, aimed at summarization, subagents and browser use. Those claims come from Anthropic’s own pages, so they are vendor claims rather than independent benchmarks. They still matter for buyers because they point in one direction: each new Claude tier does the same job for less compute. A team that tested Opus 5 in the summer and judged it too costly should rerun the numbers. Claude Code, Anthropic’s own coding agent, bills differently from Copilot. Third-party reporting says it draws on plan usage limits rather than GitHub’s per-credit metering, and that Anthropic once planned a separate credit pool for Agent SDK and claude -p usage starting June 15 before pausing the change on the scheduled date. Anthropic’s pricing page https://claude.com/pricing lists Claude Max from $100 a month with a $200 higher tier, and those Max plans include $100 to $200 a month in API credits for building on the Claude Platform. That split gives heavy users a real decision. Running Claude models through Copilot spends GitHub credits at $0.01 each. Running Claude Code directly spends a Claude subscription. Developers who want the terminal workflow can follow the Claude Code setup tutorial https://shattered.io/claude-code-tutorial-13-steps-2026/ and compare a week of real usage against the Copilot meter before committing. Codex Moves From ChatGPT Into VS Code OpenAI positions Codex as a coding-agent product spanning a desktop app, CLI, IDE extension, web and cloud workflows. Codex access is included across ChatGPT plans, including Free and Go. Codex Cloud is narrower, open to eligible Plus, Pro, Business, Enterprise, Healthcare and Education accounts depending on rollout and workspace settings. OpenAI introduced a Codex app for macOS on October 21, built to manage multiple agents and run work in parallel. The top of the ladder got an update on October 15. OpenAI announced ChatGPT Pro 500 at $500 a month with the highest included usage and access to Astra Ultrafast in ChatGPT Work and Codex. Pro now comes in three tiers at $100, $200 and $500. OpenAI also said the temporary Codex promotion on Plus would end and that Plus usage would be rebalanced toward more sessions across the week. Pricing inside Codex runs on credits per million tokens. OpenAI says usage averages roughly $100 to $200 per developer per month, with actual charges varying by model, token volume, concurrent instances, automations and fast mode. GPT-6.1 Sol sits in the Codex rate card and is rolling out to eligible paid plans. One reading of the rate card lists GPT-6.1 Sol at 50 input, 2.5 cached-input and 250 output credits per million tokens, but that line has a single source, so confirm it against OpenAI’s page before budgeting around it. Codex has also stretched beyond the editor. Code review is a growing use, and the comparison of AI code review tools https://shattered.io/ai-code-review-tools-bugbot-vs-codex-2026/ shows Codex competing with dedicated review products. Game developers get a path too, since plugins now bring Claude Code and Codex into engines, as the Unity plugin rollout https://shattered.io/unity-claude-code-codex-plugin-skills-2026/ showed. On the retirement side, OpenAI pulled GPT-5.4 and GPT-5.4 mini from ChatGPT-account Codex on August 31. OpenAI separately said on October 21 that its models and Codex agentic workflows are coming to Google Cloud Model Garden, with general availability planned for the following weeks. As of October 23 that was an announcement, not a shipped product. Cost Comparison: Where Each Agent Bills Putting the three routes next to each other exposes how differently the vendors charge. Copilot sells a credit meter with plan allowances. Anthropic sells subscriptions with API credits attached. OpenAI folds Codex into ChatGPT tiers and meters heavier work by credits per token. | Route | Where the agent runs | Entry price | Top individual tier | What the price buys | |---|---|---|---|---| | GitHub Copilot | VS Code Agents window, CLI, Copilot app, GitHub.com | $10 Pro | $100 Max | 1,500 to 20,000 AI Credits at $0.01 each | | Claude Code via Claude plans | Terminal editor integrations reported | Included with paid Claude plans | $200 Max | Max includes $100 to $200 a month in API credits | | Codex via ChatGPT | Desktop app, CLI, IDE extension, web, cloud | Included on every ChatGPT plan, Free and Go included | $500 Pro 500 | Credits per million tokens, averaging $100 to $200 per developer per month by OpenAI’s estimate | | Copilot Business or Enterprise | Same surfaces, organization managed | $19 or $39 per seat | Shared pool | Pooled credits, overage at $0.01 each | No single route wins on price. A developer who writes code by hand and leans on completions pays $10 for Copilot Pro and rarely touches the credit limit. A developer who runs several long agent sessions a day will outgrow Pro quickly. At that point the choice depends on which model the work needs, since Copilot Pro+ and above carry Opus 5.5 and GPT-6.1 Sol in one subscription, while the vendor routes tie users to one family each. Sandboxing and Credential Controls Agents that run shell commands and read repositories create a credential problem. GitHub’s answer arrived in the October 5 weekly release, which went public on October 9. Local sandboxing became generally available in Copilot CLI, the Copilot app and VS Code sessions using Agent Host. It restricts an agent’s access to files, networks and credentials, and GitHub states there is no additional charge for it. The Copilot coding agent takes a different path. GitHub describes it as taking an issue into an isolated environment powered by GitHub Actions and returning a pull request for review. Work can be assigned from GitHub.com, GitHub Mobile and VS Code, and no separate subscription is needed. Each session consumes Actions minutes and credits. Why the urgency? Repository access has become a target. The Meta GitHub breach https://shattered.io/meta-github-breach-1500-people-2026/ , which exposed code and data tied to 1,500 people, showed how much sits behind a developer token. An autonomous agent holding that token with open network access widens the exposure. Sandboxing narrows it by default, and the secret-classifier checks GitHub announced on October 7 add a second layer by scanning for leaked credentials, at the price of AI Credits. Treat the sandbox as a floor, not a guarantee. Sandboxed agents still act on whatever repository content they read, so prompt injection through a malicious issue or dependency remains a separate risk that no billing plan covers. Business and Enterprise Billing Changes Organizations face a different set of levers. On October 1, GitHub changed Business and Enterprise billing to upfront seat charging for eligible customers, effective at the start of each existing customer’s billing cycle. Additional usage, spend controls, usage tracking and credit purchases stayed available. Licenses feed a shared organization pool, and usage beyond the pool costs $0.01 per credit. Budget controls tightened in September and October. On September 18, Business and Enterprise users gained a generally available way to request a higher organizational budget after hitting their credit limit. Reportedly, those users have also been able to see current-cycle credit usage in the Copilot status menu since July 30. On October 8, GitHub added organization-level billing for Copilot code review, letting an admin charge the repository-owning organization instead of an individual member’s entitlement, with optional budgets. Reporting improved on October 22. The Copilot usage metrics API began identifying organization-created skills, custom agents, MCP servers, slash commands and plugins, where earlier output lumped them into “other” or “custom” buckets. For a finance team, that attribution turns a vague line item into something a manager can act on. One more detail affects rollout planning. GitHub began gradually enforcing a global default policy for generally available Copilot features on October 22, with the rollout reportedly continuing through October 26. Admins who rely on a feature staying off should check their policy settings before that window closes. Adoption Data From Stack Overflow’s 2025 Survey Adoption figures help explain why vendors fight over the editor. The 2025 Stack Overflow Developer Survey, the latest edition cited here, stated that “Among developers using out-of-the-box AI assistants, 67.9% reported using GitHub Copilot.” The same survey said that “Among developers using out-of-the-box AI assistants, 40.8% reported using Claude Code.” Those are 2025 numbers, taken before the autumn 2026 model wave, so they describe the starting line rather than the current race. The editor itself matters. In the survey’s wording, “Visual Studio Code was reported as the development environment used by 75.9% of survey respondents.” That share explains Microsoft’s bet on VS Code as the home for every agent. On agents specifically, the survey said that “Among respondents, 14.1% reported using AI agents at work daily, 9.0% weekly, and 7.8% monthly or infrequently.” Roughly three in ten respondents used agents at least occasionally in 2025, which left plenty of room for the 2026 push. The survey’s technology chapter, published by Stack Overflow https://survey.stackoverflow.co/2025/technology , added that “Claude Code was reported as a development environment used by 9.7% of respondents.” Put differently, Claude Code reached nearly one in ten developers in the survey year, before it landed inside Copilot. A year later, that standalone tool and the editor it competes with now share a window. How Copilot Stacks Up Against Cursor and Standalone Agents Copilot’s pitch is breadth: one subscription, many models, and the three major agent brands inside the editor most developers already use. Cursor’s pitch is depth in a purpose-built editor. The GitHub Copilot vs Cursor https://shattered.io/github-copilot-vs-cursor-4-models-8-days-2026/ comparison tracked how both vendors shipped new models within days of each other, and that race continues. Standalone agents compete on a different axis. Claude Code and Codex ship updates on a near-daily cadence, and users who want the newest agent behavior sometimes get it there before Copilot’s harness catches up. Copilot answers with consolidation: a developer can run any of the three and compare outputs in side-by-side sessions without switching tools. Teams weighing a switch can run their own test. The Cursor vs Copilot benchmark walkthrough https://shattered.io/benchmark-cursor-vs-github-copilot-2026/ lays out a repeatable harness for measuring cost per task. Anyone moving configuration across editors should read the guide on migrating Copilot instructions into Cursor rules https://shattered.io/cursor-rules-migrate-github-copilot-2026/ first, because custom instructions rarely translate one to one. The honest read: none of these products has a lasting edge on raw capability, because all of them rent the same frontier models. The differences lie in billing, workflow glue and the controls an organization gets. On those, Copilot currently offers the broadest set for enterprises, while the standalone tools offer the freshest agent features. Historical Context: From Preview to Metered Platform The route to this point took eight months. On February 4, 2026, GitHub’s changelog announced Claude and Codex in public preview on GitHub for Copilot Pro+ and Copilot Enterprise. Back then they were guests, delegated tasks from Copilot’s agent workflow. On June 1, billing moved to AI Credits, according to third-party trackers, replacing the premium-request model. GitHub’s July changelog reportedly lists Claude Opus 5 arriving in Copilot, and July 30 brought git worktree support for all three agents. September delivered the consolidation: pull requests from any agent, Dev Containers, Agent Merge in preview and Codex handoff from ChatGPT. October added the security and billing layer, with sandboxing, secret detection and organization budgets. Each step moved Copilot from a completion tool toward an agent platform that charges by consumption. The broader developer-tools category followed the same arc. Products that once sold unlimited usage now meter it, and the editor industry sits in the middle of that shift. Readers tracking the wider field can find more in the coding and developer tools coverage https://shattered.io/category/software/ . Five Predictions for AI Coding Agents 1. Credit transparency becomes a buying criterion. Expect vendors to publish per-task cost estimates, because finance teams now ask for them. Copilot’s budget alerts and the new usage metrics are early versions of that. 2. Model churn keeps accelerating. Six models left Copilot on October 19, and a steady cadence of additions and retirements looks set to continue. Pinning a model name in production prompts will carry growing risk. 3. Heavy users split their stack. Light users will stay on one subscription, while power users will pair a Copilot plan for breadth with a standalone Claude or Codex plan for depth. 4. Sandboxing turns into the default. Local sandboxing is generally available and free. Enterprise security reviews will likely demand it by policy before agents touch production repositories. 5. Credit pricing spreads further. Developer platforms outside coding assistants already price this way, as the Vercel vs Netlify credit pricing https://shattered.io/vercel-vs-netlify-pricing-comparison-2026/ comparison shows, and buyers will meet the same model in more places. What Teams Should Do This Month Start with an audit of actual usage. Pull the usage metrics, find which developers consume the most credits, and match that against plan tiers. A developer on Pro who hits the 1,500-credit limit by mid-month will pay overage at $0.01 each, and that may cost more than moving to Pro+ at $39. Next, set budgets before turning on autonomous features. Copilot’s alerts at 75%, 90% and 100% give early warnings, and organization budgets for code review cap exposure. Then check model dependencies against the October 19 retirement list and the earlier October 2 round, replacing any pinned names. Finally, turn on local sandboxing for every Agent Host session and keep the preview features automations, Agent Merge, HydraFusion to pilot groups. They are labeled preview for a reason, and none has published pricing details for multi-step runs. A two-week pilot with a handful of developers will produce better data than any vendor claim, including the ones above. Frequently Asked Questions How much does GitHub Copilot cost in October 2026? Copilot Free costs $0, Pro costs $10 a month, Pro+ costs $39 and Max costs $100. Business is $19 per assigned seat and Enterprise is $39 per assigned seat. Each paid individual plan includes a monthly allowance of AI Credits worth $15 for Pro, $70 for Pro+ and $200 for Max. What is one GitHub AI Credit worth? One credit equals $0.01. A $10 budget for extra usage covers 1,000 credits. Usage beyond an organization’s shared pool is also billed at $0.01 per credit. Do code completions use AI Credits? No. Code completions and next-edit suggestions do not consume credits on paid plans. Chat, agents, Copilot CLI and Spaces do, along with opt-in secret-classifier checks and code review. Can Copilot run Claude and Codex agents in VS Code? Yes. VS Code supports sessions for Copilot, Claude and Codex agents in its Agents window, and each can create pull requests. GitHub’s changelog dated February 4, 2026 lists the original public preview for Pro+ and Enterprise, so check your plan’s third-party agent entitlement before assuming access. Which Claude models are available in Copilot? GitHub’s changelog lists Claude Opus 5.5 for Pro+, Max, Business and Enterprise, plus Claude Sonnet 5.5 and Claude Haiku 5.5 for Pro and above. Claude Fable 5.1 reached Pro+ and higher on September 4. Plan availability varies by model. Is local agent sandboxing free? Yes. GitHub made local sandboxing generally available in Copilot CLI, the Copilot app and VS Code Agent Host sessions with no additional charge. It limits an agent’s access to files, networks and credentials. Which Copilot models were retired in October? According to GitHub’s changelog roundup, Gemini 3.7 Flash, GPT-5.5, GPT-5.4, GPT-5.4 mini, GPT-5 mini and Grok 4.5 were deprecated on October 19. An earlier round on October 2 deprecated other models across Copilot experiences. GitHub listed replacements for each.