I have been trying to get my head around the GitHub Copilot pricing changes. The old model was pretty simple. Pay $10 a month and get access to a range of models, occasionally having to wait a week or two for the latest ones. That story is gone.

Between the temporary usage guardrails already added in April and May, and the move to usage-based billing on June 1, the interesting question is no longer “what does Copilot cost?” It is “what can I actually do before the limits or the bill start getting in the way and changing my behaviour?”

That matters because GitHub is selling GitHub Copilot as a cross-model, agentic development platform now. If that is the pitch, then the important thing is not just model access. It is how much of that access is realistically usable on Pro.

What Actually Changes on June 1

The big structural shift is that GitHub is replacing premium request units with GitHub AI Credits for individual plans. One AI Credit equals $0.01 USD, and usage is calculated from token consumption, input, output, and cached tokens, at the published per-model rates.

GitHub is also splitting included usage into two parts:

• Base credits, which match the subscription price and do not change
• Flex allotment, which is additional included usage GitHub can change over time as model economics move around

This is the individual-plan picture GitHub is publishing for June 1:

So yes, there is a real included-usage uplift on paper. Pro gets $5 more than the sticker price, which is a 50% boost. Pro+ gets $31 extra, roughly 79% more than the base price. Max effectively doubles the included value.

The new pricing is more transparent, but it is also less forgiving. The old fallbacks are going away. If you run out of included usage, you either set a paid budget or you stop until the monthly reset.

Why Pro Is The Squeeze Tier

On paper, GitHub Copilot Pro still looks tempting. It remains $10 per month, the marketing page still highlights Claude and Codex access in GitHub and VS Code, and GitHub continues to bundle features such as cloud agent, code review, and unlimited inline suggestions.

The problem is that Pro now looks like the tier most likely to disappoint power users.

First, Pro no longer gets Opus models. GitHub made that explicit in the April change notice. Claude Opus 4.7 remains available in Pro+, not Pro. If the main reason you pay for GitHub Copilot is access to the strongest Anthropic models, Pro is simply the wrong plan now.

Second, the included usage is not especially large once you look at premium models and agent workflows. 1,500 AI credits sounds fine until you remember that GitHub is pricing heavy models at frontier-model rates, and agent sessions are multi-call workflows by design. A handful of large sessions can burn through that a lot faster than the old “$10 gets me Copilot” story implies. Using the GitHub usage calculator, some of my months would have gone over $100 in equivalent usage fees, and most months landed in the $30 to $40 range for personal projects outside work. That is not a huge overspend, but it is enough to make me question whether the Pro tier is still valuable for what I want from GitHub Copilot, which is keeping on top of the general behaviour of frontier models. I may need to start bouncing between OpenAI and Anthropic subscriptions directly instead.

If I were describing Pro bluntly, I would say this. It is now a decent light-to-moderate bundle for people who want autocomplete, occasional chat, and some model choice. It is no longer obviously the right plan for serious agent-heavy usage.

The New Limits Are Not Just Monthly

One of the more frustrating parts of this whole change is that GitHub has layered new guardrails on top of the old system before the June billing transition even happens.

According to GitHub’s April individual-plan change post, there are two active usage limits today for individuals:

• Session limits
• Weekly limits

Both are token-based and multiplier-aware. They are also separate from premium request entitlements. That means you can still have premium requests remaining and hit a usage limit anyway.

That distinction matters. Premium requests tell you which models you can access and how many request units you have. Session and weekly limits are separate control rails that cap how much token-heavy work you can do in a period of time.

If you hit a weekly limit, GitHub says you can continue with Auto model selection, but model choice is disabled until the weekly window resets. If you hit a session limit, you wait for the session window to reset. Pro+ gets more than 5x the limits of Pro, which tells you a lot about where GitHub thinks the pain is concentrated.

I have not found GitHub publishing a specific daily or hourly quota for individual users in the materials above. What they are explicitly documenting is:

• session limits
• weekly limits
• monthly AI-credit allowances after June 1
• generic rate-limiting language during high-usage periods

So if you are trying to model worst-case behaviour, the published guardrails are not daily and hourly buckets. They are session, weekly, and monthly, plus general rate limiting.

Annual Subscribers Get The Worst Of The Transition

If you are on a monthly plan, the June 1 transition is fairly straightforward. GitHub migrates you to usage-based billing automatically, and the new credits show up.

If you are on an annual Pro or Pro+ plan, it is much messier.

GitHub says annual plans will not auto-renew. When your annual term ends, you will be downgraded to Copilot Free unless you proactively switch. You can cancel and receive a prorated refund, or switch to a monthly paid plan and receive prorated credits for the remaining value.

The awkward part is what happens if you stay on the annual plan after June 1. You remain on premium request-based billing, but the model multipliers get worse. GitHub’s annual-plan table shows some very sharp examples:

• GPT-4.1 moves from included (0x) to 1x
• GPT-5 mini moves from included (0x) to 0.33x
• GPT-5.3-Codex moves from 1x to 6x
• GPT-5.4 moves from 1x to 6x
• GPT-5.4 mini moves from 0.33x to 6x
• Claude Sonnet 4.6 moves from 1x to 9x
• Claude Opus 4.7 moves from 15x to 27x

GitHub also notes that GPT-5.5 is temporarily available at 7.5x on annual plans, with later pricing still to be determined.

So annual subscribers miss out on the one thing that makes the June transition even vaguely palatable, the new flex allotment and the cleaner credits model. Instead, they stay on the old meter while some of the model multipliers get dramatically harsher.

If you are on annual Pro specifically, you also still miss the same Pro-only model restrictions as everyone else, which includes the loss of Opus 4.7. If you are on annual Pro+, you keep broader model access, but the economics of many of those models still get worse if you remain on the old PRU system.

This is why I think annual users are in the worst position. They are the least automatically helped by the transition and the most likely to need a deliberate decision instead of doing nothing.

So Is Copilot Still Worth Using Versus Claude Or Codex Direct?

I think the answer is yes, but only for a narrower set of reasons than it was a few months ago.

If what I want is one subscription that gives me IDE-native autocomplete, GitHub-native workflows, code review, cloud agent, CLI integration, and access to models from Anthropic, OpenAI, and Google, Copilot is still compelling. That bundle remains unusual.

Claude and Codex direct are both stronger if I mostly want a single-vendor, single-model-family relationship.

Anthropic’s own pricing is much clearer in one sense. Claude Pro is $20 monthly and includes Claude Code, while Claude Max starts at $100 monthly and offers either 5x or 20x more usage than Pro with priority access at high-traffic times. That is a direct ladder for people who know they want Claude first.

OpenAI is pushing a similar story in a different shape. ChatGPT Plus and Pro both advertise expanded Codex usage, and OpenAI also has a Business Codex plan that is explicitly pay-as-you-go. If most of your usage is really Codex, buying that relationship directly starts to make a lot more sense than renting it through Copilot and then dealing with Copilot’s extra product layers and limits.

Where Copilot still earns its keep is this:

That last row is the key one for me. If I spend most of my time in VS Code, want inline completions all day, and like being able to swap between Claude, Gemini, and Codex-style models in one place, Copilot still has a real advantage. If I already know I mainly want Claude Code or Codex, Copilot is starting to look like a middle layer that costs convenience as much as it provides it.

My Take

I think GitHub is making pricing more honest, but certainly not more attractive for Pro users.

The new Pro plan is fine for lighter use, but it is no longer the obviously good default for anyone leaning heavily on agents. Pro+ now feels more like the entry paid tier for serious use. Max exists for people who already know they are burning through these limits and do not want to think about them every week.

For annual subscribers, I would seriously consider switching rather than sitting on the request-based plan until expiry. The combination of no flex allotment, no auto-renew, a downgrade to Free at term end, and much harsher model multipliers is a pretty bad transitional deal.

And for people who are already happy living directly in Claude Code or Codex, I think the case for staying there has become stronger, not weaker.

Key Takeaways

• Copilot Pro is still cheap, but it is no longer especially generous once you account for premium models and agent-heavy workflows.
• Pro+ is where the full model story really starts, because Pro loses Opus and gets much tighter practical headroom.
• Max looks like a bigger tank, not a radically different product.
• GitHub now has three meaningful control layers for individuals: session limits, weekly limits, and monthly AI-credit allowances.
• Annual subscribers have the roughest path because they miss the flex allotment and inherit harsher PRU multipliers if they stay put.
• Copilot still wins on the bundle, especially autocomplete, GitHub workflows, and multi-model access in one place.
• Direct Claude or Codex wins when you mostly want one vendor’s best coding experience and do not need GitHub’s extra layers.

Further Resources

• GitHub Copilot is moving to usage-based billing
• Changes to GitHub Copilot Individual plans
• GitHub Copilot individual plans: Introducing flex allotments in Pro and Pro+, and a new Max plan
• Usage-based billing for individuals
• Models and pricing for GitHub Copilot
• Model multipliers for annual plans staying on request-based billing
• Preparing for your move to usage-based billing
• GitHub Copilot App