Quick lab update from February 2026: the free-tier landscape shifted again.
At the time, OpenCode Zen surfaced GLM-5 in the free roster, and GLM-5 looked competitive with Kimi K2.5 on then-current public coding leaderboard snapshots. That was useful for daily-driver experimentation, but it was not a durable claim that GLM would beat Kimi in every coding workflow.
What Was True In This Snapshot
| Metric | GLM-5 snapshot | Kimi K2.5 snapshot | How to read it now |
|---|---|---|---|
| Arena coding Elo | ~1450 range | ~1440 range | Directional leaderboard snapshot |
| SWE-bench-style coding signal | High-70s claims circulated | Mid/high-70s claims circulated | Not a current purchasing basis |
| Context | 200K | 256K | Kimi kept the context edge |
| Vision | No | Yes | Kimi kept the multimodal edge |
The important lesson was not that GLM is always ahead of Kimi. The useful lesson was that free and low-cost coding lanes can change fast, so tool recommendations need dated verification.
Current Takeaway
For current Z.AI coverage:
- Use GLM-5.1 for the model guide.
- Use Z.AI GLM Coding Plan for the commercial CTA, quotas, and referral disclosure.
- Use Smart Spend Guide for the broader upgrade logic.
For free coding paths, keep using Free Frontier Stack instead of this archived snapshot.
Related links
- /models/glm-5.1/ - Current GLM model guide
- /tools/zai/ - Z.AI Coding Plan
- /value/free-stack/ - Current free frontier AI access
- /tools/opencode/ - OpenCode Zen vs self-hosted comparison
- /value/kimi-access/ - Kimi access paths