On GitHub Copilot's new pricing

GitHub Copilot's new premium model pricing recently went into effect, and it made me realize just how good value their previous offering really was.

Last week, I began working as normal. I use Cline in VS Code, and route the API requests through the VS Code LM API. This lets me tap into all of the GitHub Copilot models I have access to through my subscription. Monday went by without a hitch, but Tuesday came, and something odd happened. I started getting errors indicating I had ran out of quota. "Hm, that's weird" I thought, given that I had never seen such a message. After digging around I found the new pricing model, which includes 300 "premium" requests per month. As I found out, Claude Sonnet 4 is one of those premium models. What quickly became quite worrying is the fact that I had ran out of premium requests in a single day. Using models in an agent loop obviously makes a lot of requests to the API, but it still is surprising that I blew through the entire quota in a single day.

With all premium models out of the picture, I was back to the non-premium models, which include GPT-4.1 and GPT-4o. For the non-premium models you get unlimited requests per month, so I tried switching to GPT-4.1, and the difference was abysmal. Frustrated, I tried other models outside of Copilot's offerings. Google offers a free tier which includes much more generous allowances, so I tried with Gemini 2.5 Pro. Compared to GPT-4.1, it was significantly better (and a much better value proposition, since its free!), but it still lagged behind Claude Sonnet 4 for my use case. In the end, I topped up my Openrouter credits, and went back to using Claude.

GitHub Copilot seemed like a steal at $10 a month, but that was considering I had unlimited access to all models. Now, even at $10, I'm struggling to see the value. If I can only get access to GPT-4.1, I might as well stick to Google's free tier, which offers a much more capable model. It doesn't matter that I get unlimited GPT-4.1 calls if I can't do any real work with it.

OpenAI's best current model, O3, which is in the same tier as Claude Sonnet 4, isn't offered in GitHub Copilot Pro, even though it's much cheaper ($2/$8 for O3 vs $3/$15 for Sonnet). It's only offered in the more expensive GitHub Copilot Pro+, at $49 a month. Even then, that plan only offers 1500 premium requests per month. In comparison, Google offers 100 daily requests to Gemini 2.5 Pro, and a further 250 daily requests to Gemini 2.5 Flash (still a very respectable model), at no cost. Combining both models, this adds up to 10500 requests per month, for free! The only issue I have found with Google's offering is the requests per minute limitation of 5 and 10, respectively, which causes timeouts in Cline, so I have to manually retry requests.

On a separate note, OpenAI does offer free tokens if you let them train on your data. At 250000 free daily tokens for O3, on top of the already cheaper offering than Claude Sonnet 4, it seems like a very cheap way to get work done, as long as you don't mind them using your data.

I don't expect GitHub to return to the previous offering. It was obvious that unlimited requests for $10 a month was unsustainable, but the current pricing doesn't seem to reflect the true pricing of models. I'll still keep my subscription, since it's included in my student plan, but I will happily use Gemini as much as I can, and O3 directly through OpenAI much more often than I did.