GitLab hosts Kimi K3, GLM 5.3 and MiniMax M3, up to 4x calls per credit
RealGeneKim · x · 2026-09-18
GitLab added three hosted open-weight models to Duo Agent Platform: Kimi K3, GLM 5.3 and MiniMax M3, claiming some match or outperform comparable frontier models in internal testing while delivering up to 4x more calls per GitLab credit.
The core argument: no single model fits every dev task, and model choice is now a price/performance optimization problem — point hard tasks at Kimi K3 or GLM 5.3, hand routine high-volume work to MiniMax M3. As agents take on longer multi-step work, cost differences compound; the metric that ultimately matters is cost per accepted change, not cost per token.
More from coding & agent
- Browser Use vs AIHawk: patched Firefox via MCP sidesteps anti-bot detection that trips Chromium agents — thefoolscard · 2026-09-18
- Free GitHub Action flags AI slop pull requests before you review them — Available_Pressure47 · 2026-09-18
- ChatGPT web can now open GitHub PRs directly, despite outdated docs saying it can't — dotey · 2026-09-18
- AuthDrift: open-source harness reproduces stale-authorization escapes in long-running agent workflows — Short-Actuary-2850 · 2026-09-18
- Sakana AI Launches Fugu Max, a Multi-Agent Orchestrator Routing Tasks to Leanest Capable Models — tkasasagi · 2026-09-18
- No Java SDK needed: one plain Java 25 file to call TypeSafe's Jev API — therealdanvega · 2026-09-18