The new Copilot pricing pushed me toward cheaper models — and likely a better workflow

[TimeLordRaps]

🏷️ Discussion Type

Product Feedback

💬 Feature/Topic Area

VS Code

Body

After the latest Copilot changes, I canceled my subscription.

Not because frontier models are useless, but because the economics no longer make sense for my workflow. GitHub’s individual Copilot prices are still listed as $10/month for Pro and $39/month for Pro+, but GitHub also introduced weekly token-based usage limits for individuals, and Claude Opus 4.7 launched in Copilot with a 7.5× premium request multiplier. GitHub also says Opus 4.7 will replace Opus 4.5 and 4.6 in the Pro+ model picker over the coming weeks. Anthropic, meanwhile, says Opus 4.7 API pricing is unchanged from Opus 4.6 at $5 per million input tokens and $25 per million output tokens. That is why this feels less like a pure upgrade and more like an effective price increase inside Copilot.
GitHub plan changes: https://github.blog/news-insights/company-news/changes-to-github-copilot-individual-plans/
GitHub Opus 4.7 in Copilot: https://github.blog/changelog/2026-04-16-claude-opus-4-7-is-generally-available/
Anthropic pricing: https://www.anthropic.com/api

My bigger takeaway is that these changes may end up pushing people toward the cheaper models anyway. GitHub described GPT-5.4 mini as fast, strong at codebase exploration, and especially effective with grep-style tools, and it launched in Copilot at a 0.33× premium request multiplier. OpenAI also positions GPT-5.4 mini as strong for coding workflows like targeted edits, codebase navigation, front-end generation, and debugging. In OpenAI’s API pricing, GPT-5.4 mini is listed at $0.75/$4.50 per million input/output tokens, while GPT-5.4 is $2.50/$15.
GitHub GPT-5.4 mini in Copilot: https://github.blog/changelog/2026-03-17-gpt-5-4-mini-is-now-generally-available-for-github-copilot/
OpenAI pricing: https://openai.com/api/pricing/
OpenAI model docs: https://platform.openai.com/docs/models

So for 99% of real work, I increasingly suspect the better decision is not “always buy the smartest model,” but “use the cheapest model that is already good enough, then make up the difference with better thinking.” Yes, Opus-class models can still matter on genuinely hard, long-horizon tasks, and Anthropic and GitHub are both claiming real gains for Opus 4.7 on multi-step and coding-heavy workflows. But that does not mean the premium is justified for ordinary app building, debugging, or iteration loops.
Anthropic Opus 4.7 announcement: https://www.anthropic.com/news/claude-opus-4-7

I’m also skeptical of the idea that bigger context windows automatically mean better real-world performance. Public long-context research still shows “lost in the middle” effects, where models degrade when relevant information is buried in the middle of long inputs, and Anthropic’s own context-engineering guidance says context should be treated as a finite resource with diminishing marginal returns. So my current view is that paying extra for more context is not automatically a win unless someone can show strong benchmarks for effective long-context reasoning, not just advertised context length.
Lost in the Middle: https://aclanthology.org/2024.tacl-1.9/
Anthropic context engineering: https://www.anthropic.com/engineering/context-engineering-for-agents

What I would really like to see is a benchmark for something closer to: how many coordinated ideas can a model actively hold, integrate, and reason over at once? I did not find a standard benchmark for that exact question. The closest public work I found measures adjacent things: long-context retrieval and reasoning under confounders, working-memory load, and multi-problem handling in a single prompt.
LongBench Pro / ICR² / related long-context eval discussion: https://arxiv.org/html/2601.02872v1

My guess is that spending 7.5× more for marginal gains is often the worse trade than spending more of your own time on first-principles thinking and better prompt engineering. Maybe the cheaper model takes 75 minutes to get you to a well-engineered result instead of 10 minutes to a rushed one. For my use case, that is the better trade.

Also, bluntly: I canceled because I am independent, unemployed, and not in a position to subsidize pricing experiments. If anyone wants to sponsor serious work on safe superintelligence, long-context evaluation, or intelligence ordinality, I am not hard to find.

[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. ⭐

[TimeLordRaps]

I will mark answered if someone provides clear evidence of models that solve the lost-in-the-middle problem, across their entire context window, with as-dense-as-possible prompt-like problem types, effectively to my understanding of it.