The ZGCM Team at Zhongguancun Academy has released ZGCM-1, a fully open-source 7B model designed to tackle competition math and agentic search.

// 1d agoMODEL RELEASE

The ZGCM Team at Zhongguancun Academy has released ZGCM-1, a fully open-source 7B model designed to tackle competition math and agentic search.

ZGCM-1 is a new 7B parameter foundation model developed by researchers at Zhongguancun Academy and the Zhongguancun Institute of Artificial Intelligence. It is designed to challenge models that are orders of magnitude larger in specific domains like mathematical reasoning and agentic search. Notably, the project embraces full open-source principles by releasing every piece of its development process, rather than just the final model weights.

// ANALYSIS

This release stands out in the crowded 7B model space by committing to complete transparency and targeting complex reasoning capabilities.

–Fully open release includes training code, data recipes, intermediate checkpoints, and experiment logs.
–Specifically optimized for competition math and agentic search, showcasing how small models can punch above their weight.
–Provides a valuable resource for the AI research community to study the full lifecycle of training a highly capable compact model.

// TAGS

aillmopen-source7bmathagentic-searchfoundation-model

DISCOVERED

1d ago

2026-09-22

PUBLISHED

1d ago

2026-09-22

RELEVANCE

8/ 10

AUTHOR

kuldeep_s_s

// KEEP READING

More AI developer news from the feed

EXPLORE FULL FEED

NEWS1h ago

Health agencies deploy Claude for Ebola response

Facing an outbreak of the rare Bundibugyo ebolavirus in the Democratic Republic of the Congo, health agencies including the WHO Regional Office for Africa (WHO AFRO), CEPI, and the DRC's National Institute of Biomedical Research (INRB) have partnered with Anthropic to deploy Claude in their frontline response. Frontline teams and researchers are using custom Claude skills and the Claude Science workbench to collapse situation report compilation to under an hour, forecast disease spread, and assemble viral genomes directly via natural language prompts.

BENCHMARK1h ago

GLM 5.3 matches frontier cyber benchmarks for less

Botsbench evaluated the newly released GLM 5.3 family from Z.ai on cybt-ctf, a 59-task SIEM investigation benchmark designed to measure agentic AI capabilities in cybersecurity environments. While matching the investigative capabilities of frontier competitors like OpenAI's models, GLM 5.3 proved most notable for its dramatic cost reduction across long-horizon autonomous workflows, positioning it as an economical alternative for production security operations.

POLICY1h ago

Amodei warns UN on AI biosecurity risks

Anthropic CEO Dario Amodei warned the United Nations Security Council of urgent dual-use threats posed by advancing AI in synthetic biology, citing Claude's recent molecular biology discoveries. To avert catastrophic misuse, Amodei urged world leaders to establish narrow international treaties banning AI-driven biological weapons, implement pre-flight safety testing, and create independent verification standards.