Please reconsider removing sonnet 3.7, copilot team!

[naina933]

Select Topic Area

Product Feedback

Copilot Feature Area

General

Body

Hey folks — new (re)subscriber here.

Support pointed me here so the Product team can see this—sharing my feedback below.

I just noticed claude sonnet 3.7 is set to be removed on Oct 23. That’s a bummer.

From my usage: sonnet 4.x is fantastic when I have a very specific ask — it’s literal and tool-centric. sonnet 3.7 has a different vibe: more proactive and exploratory, sometimes surfacing angles I didn’t spell out. That matters for conversational ideation.

Sonnet 3.7 is genuinely unique and the last model in the claude 3 series.

Please reconsider removing sonnet 3.7, copilot team!
If removal is final, I’d love to understand the rationale beyond benchmarks.

Thanks!

[github-actions[bot]]

💬 Your Product Feedback Has Been Submitted 🎉

Thank you for taking the time to share your insights with us! Your feedback is invaluable as we build a better GitHub experience for all our users.

Here's what you can expect moving forward ⏩

• Your input will be carefully reviewed and cataloged by members of our product teams. 
  • Due to the high volume of submissions, we may not always be able to provide individual responses.
  • Rest assured, your feedback will help chart our course for product improvements.
• Other users may engage with your post, sharing their own perspectives or experiences. 
• GitHub staff may reach out for further clarification or insight. 
  • We may 'Answer' your discussion if there is a current solution, workaround, or roadmap/changelog post related to the feedback.

Where to look to see what's shipping 👀

• Read the Changelog for real-time updates on the latest GitHub features, enhancements, and calls for feedback.
• Explore our Product Roadmap, which details upcoming major releases and initiatives.

What you can do in the meantime 💻

• Upvote and comment on other user feedback Discussions that resonate with you.
• Add more information at any point! Useful details include: use cases, relevant labels, desired outcomes, and any accompanying screenshots.

As a member of the GitHub community, your participation is essential. While we can't promise that every suggestion will be implemented, we want to emphasize that your feedback is instrumental in guiding our decisions and priorities.

Thank you once again for your contribution to making GitHub even better! We're grateful for your ongoing support and collaboration in shaping the future of our platform. ⭐

[sharpjonov]

Totally agree with this — I’ve noticed the same difference between Sonnet 3.7 and the newer 4.x models.
3.7 just feels more conversational and open-ended, which makes it great for brainstorming and creative work.
Would be awesome if they could keep it around or at least preserve that “vibe” in future releases.
Thanks for putting this into words — well said!

[mydear228]

Totally agree. The way 3.7 is proactive and does unprompted stuff is a bit quirky. I'm a fan of it, though.

[remimom]

Totally agree. I made my website design with Sonnet 3.7, and people still say it looks amazing.
No other model since then has been able to match 3.7’s design skills.

[robertobze]

agreed!!!

[ghost]

Personally, I'd much rather see a few high-performing models with strong personalities stick around than a bunch of lobotomized, lightweight ones.

Having more diverse and good options is always better. And I think sonnet 3.7 is a perfect example of this. It's in a great spot:

• It's only been out for a few months.
• GitHub Copilot is sourcing it from multiple vendors, which shows trust.
• The price isn't outrageous like the flagship models.
• Anthropic is guaranteeing support until Feb 2026.

The most interesting part for me is its problem-solving "personality." While 4.x models tend to take the safest, most straightforward path (which is a good trait, of course), sonnet 3.7 often explores different, more creative angles.

Unless GitHub Copilot's internal logic is something like, "Okay, so 3.7 is only a few months old, it's not that expensive, we have a steady supply from Anthropic and others, and it has a completely different vibe from 4.x... but hey, it's an 'older' model, so we have to kill it," then I see no reason for it to go away.

I love how sonnet 3.7 sometimes just takes initiative. It's a weirdly charming personality trait. I really hope to see this model around for a long time. To me, sonnet 3.7 feels very kind and cute.

[HyperBlaze456]

As well explained by @Iana174 above, deprecating Claude Sonnet 3.7 before Anthropic does feel a bit rushed.

I personally like Copilot as it supports various different models in a single unified API, making individual users easily test and experiment with different models. 3.7 Sonnet has an uniqueness that is quite distinct from any other model, even compared to its successor 4.x sonnet.

With all due respect, please reconsider the deprecation of Sonnet 3.7.

[ghost]

Oh man, so they really got rid of 3.7. I guess it shows the Copilot team's determination to immediately scrap any older model that can't code well.

Why not just line them all up by coding benchmarks, keep only the top scorer, and delete the rest? Wouldn't that be the most "efficient" way? :(

I honestly didn't think they'd actually delete such a unique model this soon, especially one that was still useful for many different things. The latest model isn't always a strictly superior in every respect, you know.

It's just... a shame.

[anirudhmsu]

I’ve had a similar experience, 3.7 seemed to explore on lot of dimensions. Would love to see that fine-tuning stay.

[parshia]

I came here when I couldn't find the 3.7 in the model list. It should not have been removed

[ItsXanadu]

It's a shame you removed Claude 3.7 - I've tried all of the models throughout my project and 3.7 was by far the best. Why was it removed? The only reason I paid for premium was because of 3.7. Man, I was on a roll!

[KateCatlin]

Hey everyone -

Thank you for all the feedback about Sonnet 3.7 being deprecated. We really do appreciate you taking the time to share your thoughts.

We know that switching models can be frustrating when they're such integral parts of all of our workflows. We recognize this is a real pain point, and we're always evaluating how we handle our model lifecycle to better balance stability with bringing you the latest improvements.

That said, based on our usage data, we made the difficult decision to deprecate Sonnet 3.7. We've put those resources toward the newer versions - Sonnet 4 and 4.5 - which have significant improvements for our most common use cases.

For those who still want to use Claude Sonnet 3.7, you can continue to do so through our BYOK (Bring Your Own Key) option with your own Anthropic API key.

We'll keep making these decisions based on usage data and performance, but we hear you on the disruption this causes. Thanks again for being vocal about what matters to you.

[naina933]

Hi, thank you for the reply, though it’s not the result I’d hoped for.

We've put those resources toward the newer versions - Sonnet 4 and 4.5 - which have significant improvements for our most common use cases.

1. From my understanding, models are hosted by providers (AWS/GCP/Anthropic). Could you share which GitHub-side resources were the constraint?

For those who still want to use Claude Sonnet 3.7, you can continue to do so through our BYOK (Bring Your Own Key) option with your own Anthropic API key.

2. My request is't just access to the model in Copilot; I want it integrated into Copilot's 'premium requests' workflow.

I respect the Copilot/Product team's decision, though I’m disappointed. Thanks for the reply.

[mydear228]

I think it would be a good idea to keep the default model list clean by hiding older models, while still offering a way to access them. An option like a "legacy slot" or "View non-recommended models" would be perfect for supporting models like Sonnet 3.7 and Opus 4.

It's a bit disappointing that the industry's leading copilot has the fewest natively supported models.

[rishavbhowmik]

@KateCatlin Is this really fair?

https://docs.github.com/en/copilot/reference/ai-models/supported-models#model-retirement-history

All the powerful models have been removed! on 2025-10-23

If I be honest, this tool just got 75% less useful for me.

[naina933]

@KateCatlin I'd appreciate it if you could take a moment to respond to my previous feedback about the 3.7 sonnet's deprecation.
I understand you might be busy, but it would be really helpful to know if this issue is being reviewed. Thank you!

[ghost]

Hey, it's been a while.

In the post "Selected Claude, OpenAI, and Gemini Copilot models are now deprecated," you asked for feedback. Dozens of people gave their feedback and opinions. But honestly, it all felt pointless.

1. What was resolved? -> Nothing.
2. What questions were answered? -> Surprisingly, none. The answers were no different from the original notification in the link.
3. Did anything change? -> Not at all.

So, what was the point of any of it? -> There was no point.

It just feels like you said, "Please give us feedback. We'll read it. We'll only read it. Just read it, lol."

Regarding the models deprecated on October 23rd: Is there absolutely no time to give us a convincing answer on this issue? If it's truly impossible to explain, there must be a reason for that as well, right?

We're looking for a genuine explanation, @KateCatlin.

[add1ct3dd]

Really frustrating to see 3.7 just disappear, you'd have thought the statistics in usage would prove that this model is worth keeping on for the forseeable. I can't get anywhere near as good results with 4.0 or 4.5, what a shame.

[ubidev]

Retiring 3.7 is a sure way as of now to make us spend more $ to get the results we want.
It needs to be reinstated until a suitable replacement is available as the current alternatives clearly aren't.

[rishavbhowmik]

I can't make any decisive claims. But I think all LLM providers are pushing for models with a smaller hot context window inside a large sparse context for a session — a large session history where a relevance algorithm (RAG) picks a small "hot" fraction of the context for the LLM to actually see on each turn.

Speculation alert: My hypothesis is that Sonnet 3.7 may have been one of the last models to use a large actual context window for each response. Which explains why it felt "proactive and exploratory" (as claimed by people in the discussion). But this approach would have higher resource overheads, which aligns with the product team's rationale.

On the business level, companies have limited infrastructure. They have 2 options:

1. Sell more powerful (and expensive) models to fewer, high-value users.
2. Get more adopters for less powerful (and cheaper-to-run) models.

And in an economy where a company's stock values are a bigger concern than the company's profits, the 2nd option is making more sense. 😊

---

Edit 12March

By "hot context window" I meant, "high-fidelity attention".

[rishavbhowmik]

In fact they got rid of all powerful models on 2025-10-23 https://docs.github.com/en/copilot/reference/ai-models/supported-models#model-retirement-history

[naina933]

Anthropic will support 3.7 through February 2026 and aws through April 2026, whereas Copilot dropped support early, which is frustrating.
Based on Claude 4.x's trajectory, Anthropic seems to be steering in the opposite direction of the 3.7 approach—unfortunate to see.
For now, all I can do is cross my fingers that the next model restores what made 3.7 work.

Appreciate everyone's input. Not the result we hoped for, but the discussion was valuable.