Unverified50% confidenceFactExact time
In the decision memo question, none of the three models fully met the standard in either of the two rounds, with a threshold of 16 points; DeepSeek produced no finished draft in both rounds, MiniMax official version wrote half a sentence in one round and scored 14 points on editing in the other, and the old Preview scored 13 and 12 points.
1
Sources
50%
Confidence
Short-term (~14 days)
Relevance
10/20/2026
First Seen
Valid until: 11/3/2026
Sources
Related Entities
Related Claims
UnverifiedThe format instruction on the last line of the prompt is the key variable determining the output quality of small models: on math tasks, after unifying the last-line format, differences in question text style caused at most a 5-point fluctuation, while changing the format line could produce a 45-point gap.65% similarUnverifiedThe limitations of this test include: limited task scale (50 tasks, 3 trials per prompt), style versions written by a single author, length confounding between the florid and normal versions, testing only English and local models below 14B, and the perplexity check covering only three smaller Qwen models and using full-precision weights while actual inference used 4-bit quantization.64% similarUnverifiedThe difference between the "9-point band" and the "7-point band" is meaningful in LLM assessment; the digits after the decimal point are not.63% similarUnverifiedOn the held-out set, the 8k–16k vocabularies had the lowest bpb across all three model sizes, while the 2k vocabulary consistently performed worst.62% similarUnverifiedOn 13 questions, the graph approach (Ratified) answered 13/13 correctly; a BM25 plus bge-small retriever on the same text hit exactly 1/10, placed within the top ten 3/10, and three counting questions could not be scored.62% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1077347API
curl https://kongchang.com/api/v1/knowledge/claims/1077347MCP
get_claim(id=1077347)