Fetching from the wire…
01
02
03
04
05
06
07
08
Models2026-07-17 · source-backed
Unisound's U2 is a 266B-total / 10B-active MoE tuned for agents, citing 72.2% SWE-bench Verified at $0.15/$0.30 per 1M tokens, while GLM-5.2 is getting named the strongest open-weight coding model across July roundups. Both are roundup-sourced, so verify the benchmarks against primary release notes before you build on them. Directionally, they're more evidence that self-hostable weights can now anchor real coding-agent work.
Each link below shares sources, entities, or timing with this story.
Shared entities / Shared topic / Earlier coverage
Both cover GLM, MoE, SWE, Verified; overlapping topics (benchmark, coding, glm-5, model); earlier GLM coverage from 2026-03-04.
Shared entities / Shared topic / Earlier coverage / Tension
Both cover GLM, MoE, SWE; overlapping topics (agent, benchmark, coding, glm-5, model); earlier GLM coverage from 2026-05-05.
Both cover GLM, MoE, SWE, Verified; overlapping topics (against, glm-5); earlier GLM coverage from 2026-03-10.
Shared entities / Shared topic / Earlier coverage
Both cover GLM, MoE, SWE, Verified; overlapping topics (benchmark, coding, model); earlier GLM coverage from 2026-06-14.
Shared entities / Shared topic / Earlier coverage / Tension
Both cover GLM, MoE, SWE; overlapping topics (benchmark, coding, glm-5, model); earlier GLM coverage from 2026-04-20.
Shared entities / Shared topic / Earlier coverage
Both cover GLM, MoE, SWE; overlapping topics (agent, benchmark, coding, glm-5, model); earlier GLM coverage from 2026-04-09.
Both cover July, MoE, SWE; overlapping topics (agent, benchmark, coding, model); earlier July coverage from 2026-07-13.
Shared entities / Shared topic
Both cover GLM, SWE, Unisound, Verified; overlapping topics (agent, model).