GitHub Copilot Coding Agent now uses one Premium Request per session

[Akash1134]

Hey Copilot Community!

We've listened to your feedback. As of July 10, 2025, the GitHub Copilot coding agent has become even more efficient with your premium requests. Now, each coding agent session uses just ONE premium request regardless of the complexity or number of files modified.

💡 What does this mean for you?

• More bang for your buck: Delegate up to 20x more tasks within your monthly premium request allowance
• Predictable usage: Know exactly what you're spending before you start a task
• Greater efficiency: Complete complex PR creations or modifications without worrying about burning through your premium requests

🔍 How does it work?

A session starts when you ask the coding agent to:

• Create a new pull request
• Make changes to an existing pull request

That's it! The entire session counts as just one premium request, no matter how complex the task.

Note: While premium request usage is now fixed, GitHub Actions minutes will still vary depending on how long Copilot needs to complete your task.

🛠️ Who can use it?

This update is available in public preview across all paid GitHub Copilot plans. For Copilot Business and Enterprise users, your administrator needs to enable the Copilot coding agent policy.

🔗 Want to learn more?

Check out the official announcement or dive into the documentation for all the details.

---

What task will you delegate to Copilot first with this new pricing model? Share your ideas below! 👇

[hayoshii]

While the new pricing model for the Coding Agent—where a full session only consumes one premium request—is a welcome improvement, it seems that Copilot Chat still consumes one premium request per message.

This can feel disproportionately expensive compared to the Coding Agent, especially when doing minor edits or having a short back-and-forth conversation, since each message still counts individually regardless of complexity.

Are there any plans to bring a session-based pricing model to Copilot Chat as well, similar to what the Coding Agent now has?

It would be great to know if such improvements are in the works.

[cmiguelangel859-del]

😎

[angelicharm]

Same here.

Github Copilot in my VS Code was a big help but I really can't practically run my projects with my multiple repos on such limited restrictions, and being that I'm not a business owner I can't see paying for Copilot Pro+ or Enterprise unfortunately, especially with so many AIs available with more reasonable restrictions and prices.

Premium Requests is confusing and very limited.

This site is confusing:
"on a paid plan, you get unlimited code completions and unlimited chat interactions...

Paid plans also receive a monthly allowance of premium requests, which can be used for advanced chat interactions."
https://docs.github.com/en/copilot/concepts/copilot-billing/understanding-and-managing-requests-in-copilot#how-do-request-allowances-work-per-plan

Github Copilot Pro has 300 Premium Requests per month total across ALL repos. Premium Requests means anything in the Copilot chat window. That is my understanding after reading that site and asking Copilot specifically.

You can have unlimited code completions in the editor window (that does not automatically check rest of repo for problems) though. How useful lol 😆.

I've been on a trial version for 2 weeks, have just 2 repos and already hit my Premium Requests cap. 🤬

[kishan-bm]

i purchased the GitHub Copilot Pro a month ago and one month over and after the one month free trial should continue but after few days the free trial has stopped and now it is asking for upgraded to Copilot pro.

[MisinformedDNA]

4.1 and 4o use standard requests and are unlimited. Claude and others are premium requests. (At least on a paid plan, starting at $10/mon.)

[angelicharm]

4.1 and 4o use standard requests and are unlimited. Claude and others are premium requests. (At least on a paid plan, starting at $10/mon.)

Thanks for clarifying 😊

[07ke]

There are still 18 days left — I feel like I should try a different place.

[AyGemuy]

wow

[cmiguelangel859-del]

Aqui Estonia tambien

[kurtzilla]

I see, we are still getting charged for fixing the agent's mistakes, which are numerous. I was hoping the new "generate instructions" would reduce the amount of botched powershell commands, bad coding practices, etc. But it doesn't seem to do any better than previous at providing valid code and commands.
It's great that issues are being addressed, but it would be nice if the marketing machine could stop touting the new features until they are ready for prime time.

[jpop104]

I want to cancel my subscription

[AyGemuy]

not bad

[Guli08]

No bad

[binidom]

Why are you deceiving people? I have been writing to your support for several days, you are selling Sonnet 3.5 under the guise of Sonnet 4. This is complete chaos, and you are silent in support

[turbotoad12]

Wow! i never knew, i just use the base gpt models. their good enough for me.

[kurtzilla]

This is very telling about the awareness of the github copilot.
Ask the agent which LLM is being used - you might not like the answer.
Guess what - we do care! How can we trust an agent that doesn't know it's origin?

(my project resides in a folder named Gemini - completely unrelated to gemini agent)

[StrayerJ]

This is a solid change.. the Agent would previously use 30-100+ premium requests when assigned a single issue; now it's 1 per session!

[kishan-bm]

Hi , I have subscribed to GitHub pro version with ₹999 so I can use any number of agent mode chats in vs code or is there any limit ?

…

On Tue, 29 Jul 2025 at 11:16 AM, StrayerJ ***@***.***> wrote: This is a solid change.. the Agent would previously use 30-100+ premium requests when assigned a single issue; now it's 1 per session! — Reply to this email directly, view it on GitHub <#165798 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/BBJTQ5Y2RQ7GDWP34DUP2H33K4DDVAVCNFSM6AAAAACBJQRK6CVHI2DSMVQWIX3LMV43URDJONRXK43TNFXW4Q3PNVWWK3TUHMYTGOJRG44DQMQ> . You are receiving this because you commented.Message ID: ***@***.***>

[MisinformedDNA]

Standard model use is unlimited. Premium model use has limits based on the subscription tier.

[Fumnanya92]

I had an active subscription for $10 and used up all my premium requests. The agent was really helpful at first, but then started making a lot of mistakes. Despite correcting them, the same errors kept happening, which ended up using more of my requests. I exhausted the premium requests and thought that when my subscription renewed, it would reset. However, since the beginning of this month, after the renewal, it hasn't reset. It keeps telling me it will reset on September 1st, which will be after my current subscription has expired. I assumed that when I pay for a new subscription, everything resets. Why am I paying this money again if that’s not the case, have I wasted my money?

[nikjayswal]

I purchased github copilot pro subscription using apple pay Rs.999 on 21-8-25 and the Subscription is ended on a single day 22-8-25 & i m using vscode Chatgpt 5 agent mode using 1x request.

i think it’s a scam or an error

[coawazie]

The agents are a complete mess. They mess up your code, overwrite unrelated things, and leave your product permanently destroyed! I have had to nuke local copies of products REPEATEDLY because copilot agent went beserk with hallucinations and inability to follow instruction or stay within specified scope. I am thinking of cancelling my subscription at this point!

[ougotti]

Could you clarify which model is used in the GitHub Copilot Coding Agent?
In particular, I’m interested in the behavior when assigning Copilot to an Issue.

Is this connected to the model selection in Copilot > Features settings, or is it independent?

I understand this is a Premium feature, but I’d also appreciate if you could explain how it relates to the pricing table — for example, which models are included under which plan.

[rahulmr2963]

Same I too want to know this.

[mfsbo]

For Copilot review agent there should be option to choose model when assigning a copilot review agent. Some PRs are not worth 1 premium request to review for example documentation spelling checks and lint/format stuff failing on PR can be done by free models but we have no choice.

[rahulmr2963]

I have GitHub Copliot Pro subscription, how can I make sure that my copilot agent always uses the GPT 4.1 model so that I am not charged for the 1 Premium Request which is required for other Models like Sonnet 4 or Sonnet 3.7