Kilo Code benchmarks GLM-5 Turbo, scores 90.5/100

Kilo Code turning itself into a model benchmarking platform is a smart move — it deepens its position as the model-agnostic alternative to Cursor while generating high-signal content for its developer audience.

• –GLM-5 Turbo scored 90.5/100 on autonomous coding tasks (API implementation, bug fixing, documentation), edging out MiniMax M2.5 at 88.5/100 — though it was twice as slow (44 min vs. 21 min)
• –Kilo made GLM-5 free on its CLI for a limited period, which is a direct growth play to funnel developers into its platform
• –The CLI itself was recently upgraded to v1.0 by forking OpenCode, giving Kilo a terminal-native path alongside the IDE extension
• –With 500+ supported models at zero markup, Kilo's BYOK model means benchmark wins for any model translate directly to user adoption
• –43% of Kilo's user base migrated from Cursor — the benchmark narrative reinforces the "same models, no lock-in" pitch