GitHub Copilot team publishes first-party data on which models perform best
unixterminal · x · 2026-11-03
Pierce Boggan of the GitHub Copilot team says they spend enormous effort optimizing the Copilot harness for the best, most token-efficient results, and are for the first time publishing their data on which models perform best inside Copilot — a first-party, harness-level comparison rather than a third-party leaderboard.
More from coding & agent
- Unofficial Crunchyroll MCP server ships 19 tools with browser-mediated sign-in — Pleasant-Today-6584 · 2026-11-03
- EAT Evolves Agent Teams to Hit SOTA on Kernel Optimization, Ranks #4 in Open Kernel Challenge — berkeley_ai · 2026-11-03
- Alchemy open-sources an agent-friendly test runner built to replace vitest — samgoodwin89 · 2026-11-03
- Alchemy adds keyless AWS↔GCP auth via OIDC Identity Federation, drops service account keys — samgoodwin89 · 2026-11-03
- Alchemy 2.0 beta ships Effect 4.1 support, DynamoDB state store with ~200x speedup, and 95+ fixes — samgoodwin89 · 2026-11-03
- Alchemy Domain ships event replay, orchestrating millions of Durable Objects via fan-out — samgoodwin89 · 2026-11-03