Brief
Open 7B model ZGCM-1 claims competitive math and agentic search results
A newly posted abstract describes ZGCM-1, a 7B open model for math and agentic search. The claims come from the authors' abstract, not from independent replication.
The abstract says ZGCM-1 is trained from scratch and uses a 256K context window. It reports that on several challenging mathematical reasoning and agentic search suites, the 7B model remains competitive with much larger frontier models, naming Qwen3-235B-A22B and GLM-5.1. It also reports a ~4.2x efficiency improvement in 16K pre-training time-to-loss. The source is an abstract, so these are the authors' reported results; no peer review or independent verification is described in the evidence.
Our reading
Our reading is that the abstract makes notable efficiency and competitiveness claims, but without peer review or independent replication, readers should treat them as preliminary.
Source details and supporting facts
Each line is stated by the page named above it.
Stated by arXiv
- ZGCM-1 is a fully open 7B dense foundation model trained from scratch.
- It supports a 256K context.
- The abstract reports a ~4.2x efficiency improvement in 16K pre-training time-to-loss.
- The abstract says ZGCM-1-7B remains competitive with frontier models orders of magnitude larger, such as Qwen3-235B-A22B and GLM-5.1, on several challenging mathematical reasoning and agentic search suites.
- The authors open-source model weights from pre-training, mid-training, and post-training stages, intermediate checkpoints, training code, per-stage data and data recipes, and W&B logs.
Sources
- arXivText stored 16 September 2026
How this story was checked. Written from the 1 page listed above, stored 16 September 2026; claims checked against that stored text on 16 September 2026.
What that means
- 5 of 5 reported statements were confirmed against the page that carries them; the rest were removed rather than published.
- Figures in the text were required to appear in the stored source text: yes. Identifiers: yes.
- The check reads stored text only: no claim rests on a fresh look that did not happen.
- Where the reporting was silent, the text says so instead of filling the gap.