GitHub Copilot changed how it bills on June 1, 2026. Premium requests are gone. You now spend AI credits at 1 credit = $0.01, consumed on token usage at each model's per-million rate. Copilot Pro is $10/mo for $15 of credits; Pro+ is $39/mo for $70; the new Max plan is $100/mo for $200. Code completions stay unlimited and never touch credits.
Claude Code bills differently. There are no per-token credits. It bundles into a Claude Pro subscription ($17/mo billed annually or $20/mo monthly), and you spend against a 5-hour rolling session window plus a weekly cap. The same subscription works across claude.ai, Claude Desktop, and Claude Code.
So the comparison is not feature-for-feature. It is a credit meter versus a time meter. Below are the exact rates, caps, install commands, and benchmark scores for both, so you can compute your own monthly cost before you commit. For Claude Code's own plan math, see Claude Code pricing.
Claude Code vs GitHub Copilot, short answer: Copilot is cheaper to start and runs more models, while Claude Code is the deeper terminal agent for teams that stay on Claude. Copilot Pro costs $10/mo and includes $15 of GitHub AI Credits (1 credit = $0.01), and code completions never spend credits. Pro+ is $39/mo for $70 of credits and Max is $100/mo for $200. Claude Code comes with Claude Pro at $17/mo billed annually or $20/mo monthly, and Max starts at $100/mo. It meters by 5-hour session windows and a weekly cap instead of tokens. Both tools now run the newest Anthropic models. Copilot lists Claude Sonnet 5.5 at $2/$10 per million tokens and Opus 5.5 at $4/$20, next to GPT-6 Sol and Gemini 3.8 Flash. Pick Copilot CLI if you want one credit meter across many models and GitHub-native delegation. Pick Claude Code if you live in the terminal on Claude models and prefer flat monthly limits.
Quick Verdict
- Pick GitHub Copilot if: you want one tool that runs Claude, GPT, and Gemini, unlimited code completions at $10/mo, and you bill by token usage in AI credits.
- Pick Claude Code if: you work in the terminal on Claude models (Sonnet 5.5, Opus 5.5), want flat time-based limits instead of a per-token meter, and want subagents and CLAUDE.md project memory.
- Run both if: Copilot covers inline completions and multi-model agent runs, Claude Code handles deep multi-file work where Opus output fidelity matters.
Copilot's June 2026 Billing Change, Explained
On June 1, 2026, GitHub switched Copilot to usage-based billing with GitHub AI Credits. The old premium request units (PRUs) are retired. One AI credit equals $0.01 USD, and credits are consumed on token usage (input, cached, and output) at each model's published per-million rate.
- Pro ($10/mo): 1,000 base + 500 flex = 1,500 credits, a $15 value.
- Pro+ ($39/mo): 3,900 base + 3,100 flex = 7,000 credits, a $70 value.
- Max ($100/mo, new): 10,000 base + 10,000 flex = 20,000 credits, a $200 value.
- Free ($0): limited chat and agent usage; GitHub says the Free plan supports Copilot CLI and agent mode.
Overage runs through a spending budget at $0.01 per credit. The important detail for cost planning: code completions and next edit suggestions are not billed in AI credits and stay unlimited on every paid plan. Credits are consumed by Copilot Chat, Copilot CLI, cloud coding agents, Copilot Spaces, Spark, and third-party agents.
Pro and Pro+ annual subscribers who stayed on legacy request-based billing after June 1 keep the old model: Pro is 300 premium requests/mo, Pro+ is 1,500/mo, extra requests are $0.04 each, and a Copilot code review costs 13 premium requests under legacy multipliers.
Pricing Side by Side
| Plan | Price | What You Get | Metering |
|---|---|---|---|
| Copilot Free | $0 | Limited chat and agent usage | Credits + completions |
| Copilot Pro | $10/mo | $15 credits + unlimited completions | AI credits ($0.01 each) |
| Copilot Pro+ | $39/mo | $70 credits ($39 base + $31 flex) | AI credits ($0.01 each) |
| Copilot Max | $100/mo | $200 credits ($100 base + $100 flex) | AI credits ($0.01 each) |
| Claude Code (Pro) | $17/mo annual or $20/mo | Includes Claude Code | 5-hour window + weekly cap |
| Claude Code (Max) | From $100/mo | 5x to 20x Pro usage, Opus access | 5-hour window + weekly cap |
Claude Pro is $17/mo billed annually ($200 up front) or $20/mo billed monthly, and it "Includes Claude Code." The free Claude.ai plan does not include Claude Code. Max starts "From $100 per month." Claude Code also runs through Amazon Bedrock, Google Vertex AI, or Microsoft Foundry on the API.
Entry-Tier Monthly Price
Lower is the cheaper door in. Copilot Pro completions are unlimited; Claude Code Pro caps by time.
Copilot meters token usage in credits beyond the included value; Claude Code meters active processing time.
Usage Limits and Rate Caps
This is where the two tools diverge most. Copilot caps you in dollars of credits. Claude Code caps you in hours of model processing.
Claude Code: time-based windows
Claude Code runs on a 5-hour rolling session window that starts at your first message, plus a weekly limit covering all models over 7 days. Usage is shared across claude.ai, Claude Desktop, and Claude Code on one subscription, with progress bars under Settings, then Usage. Reported weekly guidance (Anthropic-published ranges):
- Pro: roughly 40 to 80 hours of Sonnet per week.
- Max 5x ($100/mo): roughly 140 to 280 Sonnet hours per week.
- Max 20x ($200/mo): roughly 240 to 480 Sonnet hours, or up to about 40 Opus hours.
"Active hours" means time models are processing, not wall-clock time at your desk.
Copilot: credit-based consumption
Copilot credits are spent only by chat, CLI, cloud agents, Spaces, Spark, and third-party agents. Basic code completions never consume credits. A $10/mo Pro plan gives 1,500 credits ($15) per month; once exhausted, you pay overage at $0.01 per credit through a spending budget you control.
| Limit Type | GitHub Copilot | Claude Code |
|---|---|---|
| Meter | AI credits (token usage) | Active processing time |
| Window | Monthly credit allowance | 5-hour rolling + weekly cap |
| Completions | Unlimited, not billed | N/A (no inline completions) |
| Overage | $0.01/credit via spending budget | Wait for window reset, or upgrade tier |
| Shared across surfaces | All Copilot surfaces draw one budget | claude.ai + Desktop + Claude Code share one cap |
Per-Model Credit Rates in Copilot
Because Copilot bills on tokens, your real monthly cost depends on which model you run. These are the published per-1M-token rates that convert into credits (1 credit = $0.01):
| Model | Input / 1M | Output / 1M |
|---|---|---|
| Claude Opus 5.5 | $4 | $20 |
| Claude Opus 5 / Opus 4.8 | $5 | $25 |
| Claude Sonnet 5.5 / Sonnet 5 | $2 | $10 |
| Claude Sonnet 4.6 / Sonnet 4 | $3 | $15 |
| Claude Haiku 4.5 | $1 | $5 |
| Claude Fable 5 / Fable 5.1 | $10 | $50 |
| GPT-6.1 Sol / GPT-6 Sol (up to 272K context) | $2 | $10 |
| GPT-6 Astra (up to 272K context) | $10 | $50 |
| GPT-5.6 Sol (up to 272K context) | $4 | $20 |
| GPT-5.5 (up to 272K context) | $5 | $30 |
| GPT-5.3-Codex | $1.75 | $14 |
| Gemini 3.8 Flash | $0.75 | $3.75 |
Running Claude Opus 5.5 inside Copilot draws credits at $4/M input and $20/M output, and Sonnet 5.5 draws $2/$10. The same models in Claude Code draw against time, not tokens. If your work is bursty and short, Copilot's credit model can be cheaper. If you run long autonomous sessions on Opus daily, Claude Code's time-capped Max plan is usually the predictable choice.
Benchmarks: SWE-bench and Terminal-Bench
Copilot is model-agnostic, so it has no single agent score. Its result equals whichever model you select. Claude Code ships Claude models, so its score is fixed per Claude version.
Anthropic's Sonnet 5.5 launch numbers are the current reference for the Claude models both tools run. They are model scores reported by Anthropic, so they apply to Claude Code and to Copilot when you select the same model.
| Benchmark | Claude Sonnet 5.5 | Claude Opus 5.5 |
|---|---|---|
| Terminal-Bench 4.0 | 70.6% | 66.4% |
| CursorBench 4.0 | 55.5% | 57.8% |
| OSWorld 2.1 | 80.1% | 81.8% |
| Humanity's Last Exam (with tools) | 64.5% | 67.7% |
Haiku 4.5, the cheapest Claude model in both tools, scored 73.3% on SWE-bench Verified at its launch. Sonnet 5.5 costs half of Opus 5.5's credit rate in Copilot and posts the higher Terminal-Bench 4.0 score, so it is the default pick for most agent runs; Opus 5.5 leads on CursorBench and OSWorld.
| Metric | Claude Code | GitHub Copilot |
|---|---|---|
| Terminal-Bench 4.0 (Sonnet 5.5) | 70.6% (model score) | 70.6% if you select Sonnet 5.5 |
| Models available | Claude models only | Claude, GPT, Gemini, Grok, Kimi, MAI |
| Sonnet 5.5 cost | Inside Pro or Max time limits | $2/$10 per 1M tokens in credits |
| Code completions | None | Unlimited, not billed |
Install Commands and Requirements
Claude Code
Native install (recommended) on macOS, Linux, or WSL:
curl -fsSL https://claude.ai/install.sh | bashWindows PowerShell:
irm https://claude.ai/install.ps1 | iexOther paths: brew install --cask claude-code, winget install Anthropic.ClaudeCode, npm install -g @anthropic-ai/claude-code (Node 22 or later), and signed apt/dnf/apk repos (for example sudo apt install claude-code after adding the downloads.claude.ai repo). Native installs auto-update in the background; the stable channel runs about a week behind latest. Requirements: macOS 13.0+, Windows 10 1809+, Ubuntu 20.04+/Debian 10+/Alpine 3.19+, 4GB+ RAM. A Desktop app exists for macOS and Windows.
GitHub Copilot CLI
npm install -g @github/copilotRequires Node 22+. Also available via brew install --cask copilot-cli, winget install GitHub.Copilot, or curl -fsSL https://gh.io/copilot-install | bash. Copilot CLI supports MCP servers and the /model slash command for switching models mid-session.
Models, MCP, and Backends
Both tools support the Model Context Protocol. Claude Code adds servers with the CLI:
claude mcp add --transport http notion https://mcp.notion.com/mcpClaude Code supports http, sse (deprecated), and stdio transports, with JSON config via .mcp.json, ~/.claude.json, or claude mcp add-json. It runs Claude models only, but those Claude models are reachable through the Anthropic API, Amazon Bedrock, Google Vertex AI, or Microsoft Foundry.
Copilot is multi-model by design. One subscription runs Claude Sonnet 5.5, Opus 5.5, Haiku 4.5, and Fable 5.1, the GPT-5.x and GPT-6 families, Gemini 3.8 Flash, Grok 4.7, and Kimi K3, switchable with /model. Credits are the only thing that crosses model boundaries: pick a cheaper model and the same budget lasts longer.
GitHub Copilot CLI vs Claude Code: Agents, Delegation, Context
Both products now ship a terminal agent, which is why people search "copilot cli vs claude code" on its own. Each one reads your repository, edits files, runs shell commands, and connects to MCP servers. The differences sit around that loop.
Claude Code
Claude Code hands work to subagents you define once at project, user, managed, or plugin scope, for example a security reviewer or a test runner. Agent teams, which run several Claude Code instances with a shared task list and messages between them, are experimental and disabled by default. Project conventions live in CLAUDE.md files that load at session start.
GitHub Copilot CLI
GitHub's Copilot CLI documentation covers delegating tasks to the Copilot cloud agent, working across multiple sessions, invoking custom agents, and setting an AI credit limit. The CLI also runs programmatically and inside GitHub Actions, and it can use your own model provider. Because it bills in AI credits, switching from Opus 5.5 to Sonnet 5.5 mid-task halves the per-token rate.
| Feature | Claude Code | GitHub Copilot CLI |
|---|---|---|
| Install | curl -fsSL https://claude.ai/install.sh | bash | npm install -g @github/copilot (Node 22+) |
| Models | Claude only | Claude, GPT, Gemini, Grok, Kimi |
| 1M-token context | Depends on the Claude model | Listed for Sonnet 5.5, Opus 5.5, GPT-6 models |
| Delegation | Subagents; agent teams (experimental) | Cloud agent delegation, custom agents |
| Extensibility | MCP servers, hooks, skills, plugins | MCP servers, hooks, agent skills, plugins |
| Billing | Claude plan time windows or API tokens | GitHub AI Credits per token |
Cheaper Open-Source Model Inference for Coding Agents
Copilot's credit math and Claude Code's subscription both assume frontier proprietary models. If you want to run open-source models (DeepSeek and similar) under your own agent, where you serve them matters for both quality and price.
Morph serves DeepSeek with 16-bit (bf16) activations, no fp8 or int8 quantization. Most serverless providers quantize activations to fp8 to cut cost, which degrades output quality; keeping full 16-bit activations means responses match the reference weights. That makes Morph the best place to run DeepSeek when output fidelity matters. For coding specifically, Morph runs codegen-tuned speculative decoding plus custom low-level inference kernels built for code generation, which makes it the fastest and highest-quality option for coding agents rather than a general menu.
| Item | Detail |
|---|---|
| DeepSeek V4 Flash input | $0.141953125 / 1M tokens |
| DeepSeek V4 Flash output | $0.399625 / 1M tokens |
| Activation precision | 16-bit (bf16), no fp8/int8 quantization |
| Codegen path | Code-tuned speculative decoding + custom kernels |
See Morph Open Source Models and full pricing.
Decision Framework
| Your Priority | Best Choice | Why |
|---|---|---|
| Lowest entry price | GitHub Copilot | $10/mo Pro with $15 credits and unlimited completions |
| Predictable flat cost on heavy daily use | Claude Code | Time-capped Max plan, no per-token meter |
| Newest Claude models at a flat monthly cost | Claude Code | Sonnet 5.5 and Opus 5.5 inside Pro or Max time limits |
| Model flexibility | GitHub Copilot | Claude, GPT, Gemini, Grok, and Kimi models in one subscription via /model |
| Inline code completions | GitHub Copilot | Unlimited, never billed in credits; Claude Code has none |
| Terminal-first multi-file work | Claude Code | Reads the repo, plans, delivers reviewable diffs |
| Running open-source models cheaply | Morph Open Source Models | 16-bit DeepSeek at $0.141953125/$0.399625 per 1M, code-tuned serving |
Frequently Asked Questions
Is GitHub Copilot cheaper than Claude Code in 2026?
At the entry tier, yes. Copilot Pro is $10/mo with $15 of AI credits plus unlimited code completions that never consume credits. Claude Code needs a Claude Pro plan at $17/mo billed annually ($200 up front) or $20/mo monthly. The catch: Copilot meters token usage at $0.01 per credit, so a heavy agent user can blow past the included credits and pay overage, while Claude Code's Pro plan caps by time rather than dollars.
What changed in GitHub Copilot billing on June 1, 2026?
Premium request units were replaced by GitHub AI Credits at 1 credit = $0.01, consumed on input, cached, and output tokens at each model's per-million rate. Pro is 1,500 credits ($15), Pro+ is 7,000 ($70), and a new Max plan is 20,000 ($200). Code completions and next edit suggestions stay unlimited and are not billed in credits on any paid plan.
How do Claude Code usage limits work?
A 5-hour rolling session window starts at your first message, plus a weekly cap across all models over 7 days, shared with claude.ai and Claude Desktop. Reported guidance: Pro is roughly 40 to 80 Sonnet hours/week, Max 5x ($100/mo) is roughly 140 to 280, and Max 20x ($200/mo) is roughly 240 to 480 Sonnet hours or up to about 40 Opus hours. "Active hours" means model processing time, not time at your desk.
Which scores higher, Claude Code or GitHub Copilot?
Neither tool has one fixed score, because both run the same Anthropic models. In Anthropic's Sonnet 5.5 launch data, Sonnet 5.5 scored 70.6% on Terminal-Bench 4.0 and Opus 5.5 scored 66.4%, while Opus 5.5 led CursorBench 4.0 at 57.8% to 55.5%. Claude Code runs Claude models only; Copilot runs those plus GPT, Gemini, Grok, and Kimi models, so its result follows the model you pick.
What is the difference between Copilot CLI and Claude Code?
Both are terminal coding agents that edit files, run commands, and use MCP servers. Copilot CLI installs with npm install -g @github/copilot (Node 22 or later), runs any model in your Copilot plan, and bills chat and agent use in GitHub AI Credits. Claude Code installs with curl -fsSL https://claude.ai/install.sh | bash, runs Claude models only, and bills against Claude plan time windows or API usage.
Does GitHub Copilot include Claude models?
Yes. Copilot's model list includes Claude Sonnet 5.5 and Sonnet 5 at $2/M input and $10/M output, Opus 5.5 at $4/$20, Opus 5 and Opus 4.8 at $5/$25, Haiku 4.5 at $1/$5, and Fable 5 and 5.1 at $10/$50, all billed as AI credits at $0.01 each. Claude Code runs Claude models only.
Related Comparisons
Better Code Edits for Any Agent
Morph Fast Apply processes 10,500+ tokens/sec with high structural accuracy. Works with Claude Code, Copilot, or any AI coding tool through the API. WarpGrep adds agentic code search at $0.80 per 100K tokens.