Copilot Pro at $10/month is the cheapest way to get Anthropic, OpenAI, Google and xAI models on one plan, and it is the only one of these two tools that publishes an allowance you can do arithmetic on. Claude Code, included with Claude Pro at $20/month, is the stronger agent on hard, ambiguous work, and its usage limit is a number Anthropic has never published.
Pick Copilot if you want breadth, a budget you can model, and free unmetered autocomplete. Pick Claude Code if agentic depth is the thing you are buying and you will accept a capacity figure you cannot verify. Below: the credit arithmetic GitHub’s pricing page leaves out, and the part of each allowance the vendor reserves the right to move.
The 30-second answer
Pick GitHub Copilot if: you want the cheapest credible paid tier ($10/month), you need to stay in VS Code, Visual Studio, JetBrains or Neovim, you want several model vendors on one bill, or you want a published credit allowance you can forecast against.
Pick Claude Code if: your hard problems are multi-file refactors and long autonomous runs, you want the tool that took 46% “most-loved” in the April 2026 JetBrains survey against Copilot’s 9%, and you work terminal-first.
Run both if you already pay for either. Copilot’s completions are free of credits on every paid tier, so $30/month buys autocomplete breadth plus the best agent with almost no overlap.
Neither of these is downstream of the dispute that reshaped the category this autumn. OpenAI is terminating model supply to Cursor with a proposed 12 November 2026 cutoff — a change-of-control consequence of SpaceX’s acquisition of Cursor’s parent. GitHub’s OpenAI relationship is a Microsoft relationship and is not party to it; Claude Code never made a neutrality claim it could lose. If you are pricing an exit, start with what the Cursor migration actually costs.
Pricing
| Plan | Price | Included allowance | Token value per $1 |
|---|---|---|---|
| Copilot Free | $0 | 2,000 completions/mo + limited chat | — |
| Copilot Pro | $10/mo | 1,500 credits (1,000 base + 500 flex) | $1.50 |
| Copilot Pro+ | $39/mo | 7,000 credits (3,900 + 3,100) | $1.79 |
| Copilot Max | $100/mo | 20,000 credits (10,000 + 10,000) | $2.00 |
| Copilot Business | $19/seat/mo | 1,900 credits, pooled | $1.00 |
| Copilot Enterprise | $39/seat/mo | 3,900 credits, pooled | $1.00 |
| Claude Pro | $20/mo, $17 annual | Not published | Not computable |
| Claude Max 5x | $100/mo | Not published | Not computable |
| Claude Max 20x | $200/mo | Not published | Not computable |
| Claude Team Standard | $25/seat/mo, $20 annual | Not published | Not computable |
Prices verified 7 October 2026 against GitHub’s plan page, GitHub’s individual plan documentation, GitHub’s organisation billing documentation and claude.com/pricing. One AI credit is $0.01, so a credit allowance converts directly to a dollar ceiling on token spend. Claude Code is included on every paid Claude plan; it is not available on the free tier.
Two things fall out of that table that no vendor page states.
Capacity per dollar rises with the individual tier and collapses on the seat plans. Copilot Pro returns $1.50 of token spend per dollar of subscription, Max returns $2.00 — and Business and Enterprise return exactly $1.00. A five-person team buying five Pro seats gets 50% more token capacity for $50 than Copilot Business gives it for $95. What the seat plans buy is IP indemnity, organisation policy controls, pooled billing and a credit pool shared across the billing entity. Those are real, and on a seat plan they are the entire premium. Price them deliberately.
Claude’s column cannot be filled in. Anthropic has never published an absolute token or message figure for any plan, which is why the right-hand cells read “not computable” rather than “unlimited”. That is not a rhetorical point: on 14 September 2026 Anthropic cut weekly limits by 17% against what subscribers were using that week, describing the same change as a permanent 25% increase over a pre-May baseline. Both figures were true against different bases, and neither base was ever disclosed.
The credit-burn arithmetic
GitHub tells you the allowance and the per-model rates but never multiplies them out. Here is the multiplication.
Model a single agent session as 20 turns with a 60,000-token working context read from cache on each turn, 2,000 tokens of fresh input and 2,000 tokens of output per turn — 1.2M cached input, 40K fresh input and 40K output in total. Applied to GitHub’s published per-model rates:
| Model on Copilot | Rate (in / cached / out per M) | Cost per session | Sessions on Pro ($15) | Sessions on Max ($200) |
|---|---|---|---|---|
| Gemini 3.8 Flash | $0.75 / $0.075 / $3.75 | $0.27 | ~55 | ~740 |
| Claude Haiku 4.5 | $1.00 / $0.10 / $5.00 | $0.36 | ~41 | ~555 |
| Claude Sonnet 5.5 | $2.00 / $0.20 / $10.00 | $0.72 | ~20 | ~277 |
| GPT-6 Sol | $2.00 / $0.20 / $10.00 | $0.72 | ~20 | ~277 |
| Claude Opus 5.5 | $4.00 / $0.20 / $20.00 | $1.20 | ~12 | ~166 |
| Claude Fable 5.1 | $10.00 / $0.25 / $50.00 | $2.70 | ~5 | ~74 |
Your sessions will not be this session. The point of the table is the shape, and the shape says three things.
Copilot Pro’s $15 is roughly one Sonnet-class agent session per working day. That is a genuinely useful amount for someone whose agentic work is occasional and whose main use is autocomplete. It is not a professional agentic budget, and the gap between “one a day” and “unlimited” is where the 10-50× bill surprises reported after the June 2026 billing change came from.
Model choice moves your capacity by 10×. Gemini 3.8 Flash buys about eleven times the sessions Fable 5.1 does on the same allowance. On Claude Code the equivalent lever exists — Sonnet versus Opus, and effort level — but you are changing an unknown denominator rather than a known one. Copilot’s dropdown is a budget control with visible units.
Completions are the free part, permanently. GitHub’s documentation is explicit that code completions and next edit suggestions are not billed in AI credits and remain unlimited on all paid plans. If tab-complete is 90% of your AI use, Copilot Pro is effectively a $10 unlimited autocomplete subscription with a modest agent allowance attached, and nothing in this table binds on you.
The flex allotment, and why it matters
Copilot’s allowances are published as two numbers added together: a base allotment and a flex allotment. GitHub describes flex as an additional monthly amount on top of base credits “designed to adapt as the economics of AI evolve.”
Read the split across the ladder:
| Plan | Base | Flex | Share that is explicitly adaptive |
|---|---|---|---|
| Pro | 1,000 | 500 | 33% |
| Pro+ | 3,900 | 3,100 | 44% |
| Max | 10,000 | 10,000 | 50% |
The higher you climb, the larger the fraction of your allowance GitHub has told you in advance it may adjust. On Max, half the headline 20,000 is the adaptive half. That is worth knowing before you build a workflow whose unit economics depend on 200 credits per dollar holding.
This is the honest version of the comparison, and it does not resolve cleanly in either direction. Copilot discloses a number and flags which third to half of it is soft. Anthropic discloses no number and has moved it twice in four months. Copilot’s allowance is more forecastable and more explicitly conditional at the same time; Claude’s is neither. Both are pricing instruments rather than capacity guarantees — budget a seat accordingly, on either tool.
The shared-pool problem on Claude
Anthropic’s support documentation states plainly that Pro and Max usage limits are shared across Claude and Claude Code, so all activity in both tools counts against the same limits. Limits reset on a rolling five-hour window, with weekly caps on paid plans.
That makes Claude Pro’s $20 a single budget for everything you do with Claude. If you draft documents, research and code on the same subscription, the coding half is a residual. A heavy chat week is a thin coding week, and nothing in the interface shows you the tradeoff in advance because the total was never published. When you exhaust the allocation, Claude Code offers to continue at standard API rates — you can decline and wait for the reset, or upgrade.
Copilot has no equivalent coupling. Credits are spent on Copilot features only, and the chat assistant that shares the Copilot brand is a different product entirely.
Where Claude Code is still the better tool
None of the above makes Copilot the better agent, and the arithmetic should not be read as a verdict.
Agentic depth. Claude Code was built as an autonomous agent rather than retrofitted into one. Opus 5.5 reportedly scores 66.4% on Terminal-Bench 4.0, the benchmark closest to what a terminal agent does, and the April 2026 JetBrains survey put Claude Code at 46% most-loved against Copilot’s 9% — the largest preference gap in the category.
Orchestration. Projects in public beta splits one conversation into parallel cloud threads, each on its own branch and clone. Nothing on Copilot matches it structurally. Check Thread model and Thread effort in settings first: a new project defaults every thread to Opus at high effort, which Anthropic’s own documentation calls the setting that draws on your plan fastest.
MCP and skills. Claude Code’s extension surface — MCP servers, persistent skills, plugins — is the deepest in the category for wiring an agent into real systems.
Where Copilot wins back ground. IDE coverage is the broadest of any tool here, IP indemnity ships on the business tiers, the free tier is genuinely usable for students, and the $100 Max tier is the best measured capacity-per-dollar in the category. For high-volume mechanical work — mass renames, lint sweeps, boilerplate migrations — Copilot Max is the counterintuitively correct pick even if Claude Code stays the pick for hard problems.
Which to buy
| You are | Buy | Why |
|---|---|---|
| Mostly autocompleting, occasional agent use | Copilot Pro, $10 | Unmetered completions; ~20 Sonnet sessions a month is enough |
| Running agents daily on hard problems | Claude Pro, $20 → Max if you bind | Strongest agent; accept the unpublished allowance |
| Doing high-volume mechanical agent work | Copilot Max, $100 | $2.00 of tokens per dollar is the best rate in the category |
| Locked to JetBrains, Visual Studio or Neovim | Copilot, any tier | Claude Code is terminal-first; Copilot is where your IDE is |
| Needing several model vendors on one bill | Copilot Pro, $10 | Six vendors, one plan, published rates |
| Buying for a team that needs indemnity | Copilot Business, $19/seat | Pay the 1.0× token rate for governance, knowingly |
| Already paying for either | Add the other | $30/month total, almost no wasted overlap |
Both tools are cheaper than the decision is expensive. The variable that actually moves your bill is which model you select and how much context your agent keeps resident — and on only one of these two can you see that variable in units.
Next: Copilot vs Cursor if the editor is the question, Cursor vs Claude Code for the comparison most readers land on next, and Claude Code vs Antigravity if an offline option matters. Full reviews: GitHub Copilot and Claude Code. Wider field: best AI coding tools, best AI harnesses, and the developer tooling guide. To price a specific workload in tokens rather than seats, use the API cost calculator.
What changed
- 7 Oct 2026: Page created. Copilot individual and seat allowances, per-model Copilot token rates and Claude plan prices verified against vendor documentation; capacity-per-dollar and flex-share arithmetic derived from GitHub’s published base/flex split.
GitHub Copilot vs Claude Code — frequently asked questions
Is GitHub Copilot or Claude Code better value at the entry price?
They answer different questions. Copilot Pro is $10/month and includes 1,500 AI credits — $15 of token spend at GitHub's published per-model rates, plus unlimited unmetered code completions. Claude Code is included with Claude Pro at $20/month ($17 billed annually) and publishes no token or request allowance at all. If you want a number you can budget against, Copilot is the only one of the two that gives you one. If you want the strongest agent on hard problems, that is Claude Code, and you accept an allowance you cannot compute.
How many agent sessions does Copilot Pro's $15 of credits actually buy?
On our modelled session — 20 turns, a 60K-token context read from cache each turn, 2K fresh input and 2K output per turn — Copilot Pro's allowance covers roughly 20 sessions a month on Claude Sonnet 5.5, about 12 on Claude Opus 5.5, and about 5 on Claude Fable 5.1. That is roughly one Sonnet session per working day. Code completions and next edit suggestions do not draw on credits at all, so a developer who mostly tab-completes will never spend the allowance.
Does Claude Code use the same usage limit as the Claude chat app?
Yes, and this is the fact most buyers miss. Anthropic's own support documentation states that Pro and Max usage limits are shared across Claude and Claude Code, so all activity in both tools counts against the same pool. If you also use claude.ai for writing, research or document work, your coding capacity is whatever is left over. Copilot's credits are spent on coding only.
Which one gives you more model choice?
Copilot, decisively. One Copilot plan — including the $10 Pro tier — carries Anthropic Claude, OpenAI GPT, Google Gemini, xAI Grok, Microsoft MAI-Code and Moonshot Kimi models, and every plan can access every listed model. Claude Code is Anthropic-only by design. That is a deliberate product choice rather than a limitation, but it means your tool and your model supplier are the same company.
Is Copilot Business better value than several individual Pro seats?
Not on token value. Copilot Business is $19 per seat for 1,900 credits and Enterprise is $39 for 3,900 — exactly 100 credits, or $1 of token spend, per dollar of seat price. Individual Pro returns 150 credits per dollar and Max returns 200. You pay the premium on seat plans for IP indemnity, policy controls and centralised billing, not for capacity. Price those governance features explicitly rather than assuming the business tier is the cheaper one.
Can I run both?
Yes, and it is the most common professional setup. Copilot's unmetered completions cost nothing extra once you pay for any tier, so keeping it in your IDE for autocomplete while running Claude Code in the terminal for agentic work costs $30/month combined and wastes very little. The overlap is in agent mode, which you would simply stop using on whichever tool you did not pick for hard work.