GitHub Universe
24 tasks
of real work on ten setups. Each bar is tasks finished; the bright ones ran locally
San Francisco, 28 October 2026
When is local Copilot worth it? Three open-weight coders vs. Copilot out of the box.
The same Copilot CLI agent on the same real tasks: open-weight models you serve yourself against hosted Copilot.
The talk
Bring your own key for GitHub Copilot is now generally available, and open-weight coding models keep landing on Hugging Face every few weeks. But it does not cover code completions, embeddings or repository indexing. So when does running a local model actually pay off?
The session compares three open-weight coders on a single RTX 4090 and an Apple M5 against Copilot's default agent mode on real repository tasks, and shows how to run the same comparison on your own repositories.
The evidence
On 24 paired tasks, Qwen3.6 on the RTX 4090 finished 42% and Copilot Pro+ on Auto finished 92%. Qwen3.6 ran with reasoning off while the hosted models ran as GitHub serves them, which may understate it. The study compares whole configurations from one run per task, and it supports no plan ranking or buying rule.
42%
of 24 real tasks with Qwen3.6 on an RTX 4090, against 92% on Copilot Pro+ Auto
GGUF, October 2026
When is local Copilot worth it?
The same Copilot CLI agent on the same real tasks: open-weight GGUF models on an RTX 4090 and a MacBook Air M5, against hosted Copilot.
Where and when
- Conference
- GitHub Universe, San Francisco
- Date
- Wednesday 28 October 2026, 4:25 PM to 4:40 PM PDT
- Room
- Demos & Donuts, Gateway Pavilion, Level 2
- Session
- BLD1241D, a product demo, in person
- Speaker
- Alpha Romer Coma