GitHub Copilot Just Got a $100 Plan. Here's What Each Tier Actually Buys You.

GitHub announced today that the $10 Copilot Pro plan now includes $15 in effective monthly credits, and dropped a $100/month Max tier with $200 in effective credits on the same day. The math is not as simple as it sounds: a single afternoon of agentic coding on GPT-5.5 can drain a Pro account before dinner.

As of June 1, GitHub is moving Copilot to usage-based billing, meaning you don’t rent a seat anymore—you buy a monthly credit allowance and burn it on whatever models and workloads you choose. For the first time, Copilot pricing becomes transparent and comparable. We ran the numbers. Here’s what each tier actually buys.

The New Lineup at a Glance

According to the GitHub announcement, all paid tiers unlock unlimited code completions and next-edit suggestions. The paid difference is model access and agentic features. That matters.

Base vs. Flex: What Actually Happens When You Run Out

Base credits are locked. They match your subscription price and never change. Flex allotments sit on top—they’re the variable layer. GitHub designed flex allotments to adapt as model pricing shifts and new models land. You hit your base limit, the flex kicks in automatically. Hit both, you can set an additional budget to keep going, or you’re cut off.

One critical detail: code completions don’t consume credits on any paid plan. Autocomplete is the moat GitHub is protecting. Chat, agents, and multi-model workloads burn your budget. Autocomplete doesn’t.

The Model Rate Card You Need to Bookmark

This is where the actual cost lives. Each model has a per-token rate, converted to credits (1 credit = $0.01).

All rates are from GitHub’s models and pricing documentation. GPT-5.5 is the most expensive; Grok is the cheapest. Your choice of model determines how fast your credits evaporate.

The Pro Math: $15 a Month

Let’s run a real scenario. You fire up a 30-minute agentic coding session on GPT-5.5—think: multi-file refactor, agent looping through your codebase, iterating on a bug.

Rough token burn: 50K input, 20K output. At GPT-5.5 rates, that’s about 86 credits ($0.86). With $15 total, you get roughly 17 full sessions per month. That’s less than one heavy workday per week if you’re hammering frontier models. If you stick to Grok or cached-context layers, you stretch further. But agentic work on GPT-5.5 is expensive.

Who Pro is for: Autocomplete-primary devs who touch chat maybe once a day, usually on simpler tasks. Light experimentation. Not sustained agent work.

Pro+ and Max: The Real Power Tiers

Pro+ ($70 effective) is 4.6x the credits. Max ($200 effective) is 13.3x.

At Pro+ ($70), that same 17 heavy sessions becomes 81 sessions per month. You can run daily agentic work on frontier models without rationing. Max ($200) gets you to 232 sessions—essentially unlimited for solo work, or comfortable splitting between two devs on a small team.

Claude Code vs. Copilot has its own calculus because Claude’s per-month pricing is different. But within Copilot: Pro+ is the breakpoint where daily power users stop feeling the meter. Max is for teams or heavy agent-user solo devs who need all-day Opus or GPT-5.5 access without rate-limiting themselves.

The Cursor Comparison Just Changed

This matters if you’re evaluating Cursor vs. other IDE subscriptions. Cursor’s Individual plan is $20/month and advertises frontier model access. Windsurf is similar. The difference used to be opaque—different billing models, different per-request costs.

Now that Copilot’s pricing is transparent and metered the same way, you can actually compare value. Copilot Pro delivers $1.50 of effective credit value per dollar paid ($10 buys $15 in credits). Cursor advertises “frontier model access” without publishing a token-level rate card, which makes a direct credit-for-credit comparison impossible. What we can say: if you want model-level cost transparency, Copilot’s new billing is the cleaner system.

If you want Opus inside Copilot, you’re burning expensive credits. If you’re happy with GPT-5.5 or Grok, Copilot’s per-credit cost wins. Check our full Cursor breakdown for the IDE-level comparison, or the broader May pricing context if you’re cross-shopping IDEs.

What This Means: The Verdict

• Free: Limited completions. Autocomplete only. Fine for learning.
• Pro ($10): Autocomplete power users, light chat. Heavy agentic work will drain you. Refresh cycle: maybe once every two weeks of steady use.
• Pro+ ($39): Daily power users. Agentic work is sustainable if you’re not exclusively on GPT-5.5. Standard tier for professional development.
• Max ($100): All-day agent access, no rationing. Teams or solo devs burning Opus 24/7. The price of entry for sustained high-volume work.

The switch to usage-based billing is GitHub’s honest move. You see what you’re actually spending. Abuse it, you pay more. Use it lean, you save. That’s cleaner than seat licensing pretending everyone uses the same thing.

One last thing: unlimited code completions stay unlimited on all paid plans. If you’re primarily using Copilot as autocomplete—the thing it does best—even Pro covers you. The credit burn happens when you layer agents and multi-model chats on top. Choose your tier based on that, not on hype.