Unverified50% confidenceBenchmarkExact time
On the NL2RepoBench benchmark, QwenGyre raised the score from 52.5% to 58.5% in 48 steps, and end-to-end training time was shortened by about 44% compared with the Async scheme.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/12/2026
First Seen
Valid until: 1/10/2027
Sources
Related Entities
Related Claims
Unverified3,189 successful rollouts were collected from Qwen3.8-27B for supervised fine-tuning (imitation learning), and performance stagnated at 47.5%, below the results of two rounds of RL training.68% similarUnverifiedExperience internalization fine-tuning improved Qwen3-8B-27B's one-shot score from 37.07 to 48.22, while reducing generated tokens from 6.41 million to 570,000 (an 11-fold reduction).66% similarUnverifiedNCP-ArchPreview achieves the final pretraining loss of OLMo-3-7B while consuming only 51.3% of the training tokens.62% similarUnverifiedIn the LongMemEval benchmark, BM25 accuracy improved from 78.8% to 85.2% with 5 chunks.61% similarUnverifiedIn the Terminal Bench test, Qwen 3.6 27B scored approximately 60.7, while Muse Glimmer scored only 51.7, falling significantly behind.61% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1006625API
curl https://kongchang.com/api/v1/knowledge/claims/1006625MCP
get_claim(id=1006625)