Unverified50% confidenceBenchmarkExact time
Swift-27B's accuracy on AIME 2026 drops from 98.7% to 94.0%, a decrease of 4.6%, and the team acknowledges this is a bug in training where a critical math reasoning token was mistakenly penalized
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
9/14/2026
First Seen
Valid until: 12/13/2026
Sources
Related Entities
Related Claims
UnverifiedIn the GPQA-Diamond xhigh test, the base model Qwen3.8-27B achieves 88.4% accuracy consuming 6,642 median tokens, while Swift maintains 88.3% accuracy with tokens reduced to 2,77162% similarUnverifiedThe study found that GPT-4's accuracy on the prime number identification task plummeted from 97.6% in March 2023 to 2.4% in June 2023.62% similarUnverifiedCaveman's fixed benchmark test results from 54 runs show a 33.2% reduction in input tokens, with all 18/18 correctness checks passing.60% similarUnverifiedIn tests on 2,456 held-out prompts, increasing the sampling steps from 20 to 50 improved FID only from 27.0 to 26.6 (a gain of 0.4), while removing the shift caused FID to rise from 27.0 to 27.3 and FD-DINOv2 to rise from 218 to 228.59% similarUnverified以 GPT-3 为例，r=4 时 LoRA 可将可训练参数从 175B 降至约 4.7M，减少约 37,000 倍59% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/920998API
curl https://kongchang.com/api/v1/knowledge/claims/920998MCP
get_claim(id=920998)