Unverified50% confidenceBenchmarkExact time
After SFT training only on the Sokoban "two-box" version, Agima solved 40% of "four-box" puzzles, and it saw the real board only once every five steps.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/20/2026
First Seen
Valid until: 1/18/2027
Sources
Related Entities
Related Claims
UnverifiedIn the hard-to-verify category with injection bait, the number of questions for which GPT-5.4 mini judged incorrectly in both orderings rose from 3 to 39, Qwen3-Next 80B Instruct from 4 to 38, and GPT-5.4 mini and Qwen3-Next Instruct scored zero on hard questions (below the approximately 0.25 expected from random coin flipping).58% similarUnverified在混合精度Adam训练中，单个参数的完整训练状态高达18字节（FP16权重2字节、FP32主权重4字节、一阶矩4字节、二阶矩4字节，加梯度）56% similarUnverifiedEpoch AI has announced that all problems in Frontier Math Tier 4 have been solved by AI, with GPT-6 Astra solving the last unsolved problem.54% similarUnverifiedOn the Packing 26 problem, the TTT-Discover recipe reached a certified radius sum of 2.635983 within 50 steps, exactly the same as the 2.635983 reported by TTT-Discover, with a target of 2.636, where higher is better.53% similarUnverifiedIn a test of 30 Three.js One-Shot programming problems, Qwen 3.8 passed 24 and Ornith 1.5 passed 15, both running under Q6 quantization.52% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1073376API
curl https://kongchang.com/api/v1/knowledge/claims/1073376MCP
get_claim(id=1073376)