It’s easy to forget that GitHub Copilot appeared on the scene a year and a half before ChatGPT. The AI assistant has grown alongside LLM innovations, but is currently losing trust among developers. The lifeline: the competition is also feeling the pressure of rising costs, with the end of subsidized AI costs in sight.
Anyone who describes LLMs as a “complex autocomplete” can cite GitHub Copilot (GHCP) as a textbook example. When it was announced as an “AI pair programmer,” its functionality revolved primarily around completing code as you write. GitHub cited Codex, an advanced variant of OpenAI’s GPT-3 model, as the power source behind the tool. Thanks to the collaboration between Microsoft and OpenAI, developers were later able to use GPT-3.5 (the LLM behind ChatGPT when it was released), GPT-4, and increasingly advanced models from OpenAI thereafter. Now, certain paying users can also choose Claude, and Google Gemini and proprietary LLMs are also available via APIs.
GitHub Copilot’s Many Rivals
In the meantime, some rivals have emerged and already disappeared. Devin, once introduced as the “first AI software engineer,” seems to have vanished—though there are rumors that its creator, Cognition AI, might be closing a funding round with a $25 billion valuation. But Claude Code, Windsurf, Cursor, and many other AI-driven solutions now occupy a more prominent (and positive) position in developer discussions than GHCP.
What’s striking is that the “mindshare” surrounding these AI tools doesn’t match their adoption at all. According to a JetBrains survey, GitHub Copilot is used by 29 percent of developers worldwide. The Net Promoter Score (NPS) for GHCP is 11, which is positive but significantly lower than that of Claude Code (54), Cursor (34), and Google Antigravity (32).
Adoption of this developer tooling is harder to measure than it seems. For instance, 90 percent of Fortune 100 companies use GHCP; by mid-2025, it already had 4.7 million paid subscribers and was growing at a 75 percent annual rate toward 20 million users at the time. The figures for Claude (and Claude Code specifically) are harder to come by. Still, the trend is positive for Anthropic, and similar growth for GHCP is something GitHub or Microsoft would be more than happy to highlight.
Pricing Issues
External influences aside: GitHub itself is contributing to its own lack of appeal. After years of subsidized costs and a “flat fee,” the price for GHCP is now usage-based. Copilot usage consumes GitHub AI Credits. The rationale for this policy change is simple: GHCP is much more than it ever was. According to GitHub, it is now an agentic platform, with agentic usage as the standard. Because the pressure on GitHub’s infrastructure is significantly higher for long-running agentic workflows than for chat sessions, pricing must be adjusted accordingly. Those who still use GHCP in the lightweight way they always have could, in principle, pay less than before. Whereas Premium Request Units previously restricted access to the best AI models, this has also been a thing of the past since June 1.
The models within GHCP have also seen significant price increases. According to the documentation, anyone wanting to run Claude on this platform must pay about nine times more than was previously the case. Other models are expected to see less steep price increases on average, but will still become significantly more expensive. The trend shows that GitHub itself is not the only one to have concluded that the subsidized era for AI usage has come to an end. Anthropic, on the one hand a major rival to GitHub and on the other a provider via Copilot, has also realized this and is leveraging its position as a recognized quality leader for coding-specific LLMs.
The conclusion is that GitHub Copilot enjoyed a temporary appeal that is not sustainable in the long term, even for other AI tools. Since all AI rivals are currently pushing prices up roughly at the same time, migrating away from GitHub Copilot is not as obvious as one might think in isolation. The question remains, however, whether “real” AI prices are actually affordable. General AI adoption, not market share among AI tools, will serve as the benchmark for this.
Read also: AI pressure forces GitHub to take action on Copilot offering