Unverified50% confidenceFactExact time
Preliminary endpoint results on the 32B model show the same direction: 35.53% for SFT and 22.37% for final RLVR.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/23/2026
First Seen
Valid until: 1/21/2027
Sources
Related Entities
Related Claims
UnverifiedOn the HLE benchmark, COMED improved GPT-5.5's performance from 23.1% to 28.1%.75% similarUnverifiedUnder an 18k character budget, anchored hybrid retrieval achieved 36.18% accuracy, BM25 achieved 35.98%, while the best direct planner-guided method reached only 34.19%.69% similarUnverifiedOn pure reasoning tasks, Qwen3 27B scores only 30.8 on HLE, notably behind Opus's score of 40.0.68% 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 228.68% similarUnverifiedRL can let a 4B model close more than half the gap to a 27B model, but the gains have a clear ceiling.67% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1088121API
curl https://kongchang.com/api/v1/knowledge/claims/1088121MCP
get_claim(id=1088121)