On March 12, 2026, GitHub quietly did something remarkable: they removed premium AI model access from their free Copilot Student plan. GPT-5.4, Claude Opus 4.6, and Claude Sonnet 4.6 — gone. The announcement post got 2,874 downvotes and 21 upvotes. Microsoft’s response? “Pay up.”
This is a failure. Not just for students, but for the entire premise of making AI tools accessible to the next generation of developers.
What Actually Happened #
GitHub moved the free Copilot Student plan to a new “GitHub Copilot Student” tier and surgically removed access to the most capable models. Starting March 12, students can no longer manually select:
- GPT-5.4 — OpenAI’s top coding model
- Claude Opus 4.6 — Anthropic’s most capable model
- Claude Sonnet 4.6 — The previous go-to for complex coding tasks
What they replaced it with: Claude 4.5 Haiku, Gemini 3.1 Pro, and GPT-5.3 Codex. Cheaper models. Significantly less capable.
GitHub’s VP of Developer Relations, Martin Woodward, characterized it as “making student access sustainable.” The community responded with near-unanimous backlash.
The exact vote ratio — 2,874 downvotes to 21 upvotes — tells you everything about how students feel about this “sustainability” argument.
Why This Matters for Students #
Students aren’t asking for premium models as a luxury. They’re asking for them because the most capable models are demonstrably better at:
- Explaining complex coding concepts step-by-step
- Debugging intricate production issues
- Handling large-scale refactoring
- Teaching advanced software engineering patterns
As one student put it in the GitHub forum:
“For many of us working on advanced engineering projects, Claude 4.6 Sonnet and Opus are not just ‘options’ — they are currently the most capable AI agents for coding, logic, and handling large-scale refactoring. Restricting these models from self-selection limits our ability to learn with the industry’s leading technology.”
Another:
“The removal of premium models makes learning programming more difficult. These models are much better at explaining complex coding concepts, helping debug problems, and guiding students step by step.”
GitHub’s response to these concerns? A suggestion to upgrade to Copilot Pro or Pro+ — while retaining other Student Pack benefits. In other words: pay us money.
Microsoft and GitHub: Prioritizing Profit Over Education #
Let’s call this what it is. Microsoft is cutting costs by removing expensive API calls from the free tier. The “sustainability” framing is corporate speak for “we’d rather not subsidize this.”
The Register pointed out the uncomfortable context: only 3.3% of Microsoft 365 and Office 365 customers actually pay for Copilot Chat. Investors are nervous about AI capex. So GitHub — owned by Microsoft — is reducing exposure by cutting off students.
This is short-term thinking dressed up as long-term strategy. Students who get access to good AI tools today become power users, advocates, and eventually paying customers. Making their experience worse now is how you breed resentment and drive people to alternatives.
GitHub’s CEO Thomas Dohmke even said in 2024 that “students are the most important community for the future of AI-assisted development.” Actions speak louder than blog posts.
The Silver Lining: Cheap Chinese Models Are Actually Good Now #
Here’s where it gets interesting.
On March 18, 2026, MiniMax M2.7 launched scoring 56.22% on SWE-Pro — close to Claude Opus 4.6’s performance. Kilo Code ran both models through three identical coding tests:
- Full-Stack Event Processing System (35 points)
- Bug Investigation from Symptoms (30 points)
- Security Audit (35 points)
Results:
| MiniMax M2.7 | Claude Opus 4.6 | |
|---|---|---|
| Bugs Found | 6/6 | 6/6 |
| Security Vulns Found | 10/10 | 10/10 |
| Fix Quality | Good | Excellent |
| Test Coverage | 20 unit tests | 41 integration tests |
| Total Cost | $0.27 | $3.67 |
MiniMax M2.7 found every single bug and every single vulnerability that Claude Opus 4.6 did. The detection rate was identical. The cost was 14x lower.
This isn’t a fluke. MiniMax M2.5 — the previous version — is already the #1 most-used model across all Kilo Code modes, ahead of Claude Opus 4.6, GLM-5, and GPT-5.4. In Code mode, it accounts for 37% of all usage. In Ask mode, 35%.
Students locked out of GitHub Copilot’s premium models have real alternatives that don’t cost $20/month.
Free Alternatives That Don’t Suck #
Kilo Code (kilocode.ai) #
Kilo Code is a free, open-source AI coding assistant for VS Code and JetBrains IDEs with 1.5 million+ users. It offers:
- MiniMax M2.5 — Currently the #1 most-used model on the platform, scoring 54.56% on SWE-Pro
- Claude 4.5 Haiku — Fast, capable, free
- GPT-5.3 Codex — Solid performer
- Auto mode — Platform picks the best model for your task
The platform is completely free. No subscription. No Copilot required.
OpenCode #
OpenCode is another free coding agent that works with multiple providers:
- Big Pickle — Fast, free model good for quick tasks
- Mimo v2 Pro / Omni — Free tiers available
- Connects to Ollama for local models
Ollama (Local Models) #
If you have a decent GPU or just want to experiment:
- qwen2.5-coder — Alibaba’s coding-focused model, free and locally runnable
- deepseek-coder-v2 — Another strong open-weight coding model
- llama3.1 — Meta’s general model with coding ability
These run entirely on your machine. No API costs. No internet required.
What This Means for Students #
GitHub and Microsoft failed students here. There’s no polite way to spin removing access to the industry’s best coding models from a free educational tier as anything other than a cost-cutting move that hurts learners.
But the landscape has shifted in an unexpected way. Chinese AI labs like MiniMax, Qwen, and DeepSeek are releasing capable models at a fraction of the cost of their Western counterparts. Detection rates are matching frontier models. The quality gap — while real in areas like fix thoroughness and test coverage — is closing fast.
Students today have options. Kilo Code, OpenCode, Ollama, and other free tools mean you don’t need GitHub Copilot’s premium tier to get good AI-assisted coding help.
Is it ideal? No. A student with free access to Claude Opus 4.6 through Copilot has a better experience than one relying on MiniMax M2.5 Haiku. The context windows are smaller, the test coverage is thinner, the fixes aren’t as polished.
But “not as good as the premium option” is a far cry from “no option.” The tools exist. They’re free. They work.
The Bottom Line #
GitHub cut student access to premium AI models because it was expensive and Microsoft needed to show investors they were controlling AI costs. That decision hurts students, signals that GitHub’s “education first” rhetoric was always hollow, and will likely accelerate adoption of free alternatives that don’t require a subscription.
The irony: those same alternatives are often powered by Chinese models that are increasingly competitive. MiniMax M2.5 is already the most-used model on Kilo Code. Open-source Chinese AI is quietly becoming the equalizer that Microsoft’s paywall tried to withhold.
If you’re a student affected by this: try Kilo Code. Try OpenCode. Try Ollama. The gap between free and premium is smaller than GitHub wants you to believe.
And maybe — just maybe — that’s the push the open-source AI ecosystem needed.
Have thoughts on GitHub’s decision? Reach out — I’m always down to talk about AI, coding tools, and the ongoing mess that is AI pricing for students.