Open weight AI: An antidote to GitHub Copilot bill shock

The bill shock arrives

On 1 June, GitHub Copilot moved from a flat seat fee to usage-based billing [1]. Since then, agentic developers have reported bills significantly higher than their previous plans, with some reporting costs more than 10x greater than before [2].

There is also no automatic safety net. GitHub's previous fallback to cheaper models is gone, and spend is uncapped unless organisations explicitly opt into usage limits. Those limits can help control costs, but they can also restrict developers when they need AI the most.

The question is: can you reduce costs without sacrificing capability?

Open weight models as an alternative.

One option many teams overlook is GitHub Copilot's new Bring Your Own Key (BYOK) capability [3].

BYOK allows developers to connect Copilot to models outside GitHub's bundled pricing structure. One such model is GLM 5.2, a recently released open-weight model that has attracted significant attention [4][5].

The chart below shows results from the Artificial Analysis Intelligence Index [6]. GLM 5.2 scores competitively against leading proprietary models.

While benchmarks never tell the whole story, they suggest organisations can explore lower-cost models without taking a major step backwards in capability.

The Economics

The pricing comparison is shown below (prices in AUD), as per the GitHub’s pricing tables [8]:

If we look at output token costs and compare GLM 5.2 against the average cost of the other leading models, GLM comes out more than 3x cheaper.

Put another way, for every $1,000 currently spent on output tokens, the equivalent workload could cost less than $280 using open weight LLMs.

For teams embracing agentic development, where token consumption can grow rapidly, these savings can materially improve the economics of AI-assisted software development.

It is easy to try

The good news is that testing this requires very little effort.

Using ResetData's Sovereign AI Platform [9], developers can generate an API key and connect it to GitHub Copilot using BYOK [3]. The process takes around ten minutes:

Sign up to the ResetData AI Platform and generate an API key (new users receive $50 of credits).
Follow GitHub's BYOK guide [7] or use the setup image reference below.
Compare quality, developer experience and costs for yourself.

As an added benefit, ResetData's platform runs on sovereign Australian infrastructure, allowing organisations to explore lower AI costs while keeping workloads onshore.

As AI coding assistants become increasingly consumption-based, organisations owe it to themselves to test the alternatives. With GitHub Copilot's BYOK support and ResetData's Sovereign AI Platform, that comparison has never been easier; try it on one team for a few weeks, compare the bill, switch back if you don't like it!