Unverified50% confidenceBenchmarkExact time
The trained 4B-parameter model outperforms all 7B open-source math-specialized models evaluated on the counterexample generation task and remains competitive with six frontier commercial APIs.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/5/2026
First Seen
Valid until: 1/3/2027
Sources
Related Entities
Related Claims
UnverifiedA base model is the original version that has completed pre-training on large-scale data but has not been fine-tuned for a specific task, retaining the most complete parameter space and generalization potential.72% similarUnverifiedPerforming task-specific reinforcement learning on a small 4B-parameter model improved its mapping reasoning accuracy by 15.7 percentage points.71% similarVerifiedFor niche domains with insufficient training data, large models tend toward hallucination-based completion, generating API calls, parameter names, or return structures that appear plausible but are actually incorrect69% similarUnverifiedThe choice of a 4B-scale base model is intended to strike a balance between accommodating complex classification logic and controlling the computational overhead of fine-tuning inference.69% similarUnverified使用Gemma 4系列模型验证了模型权重与Harness状态的联合学习效果，发现单独的监督微调和离线强化学习预热都不能让模型取得里程碑进展，只有联合学习循环才能产生持续提升69% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/975477API
curl https://kongchang.com/api/v1/knowledge/claims/975477MCP
get_claim(id=975477)