Unverified50% confidenceBenchmarkExact time
On the Qwen3-VL series of models (from 2B to 30B-A3B) and on InternVL3.5-8B, SPLIT-RL improves average accuracy by 1.4 to 6.1 percentage points compared to standard GRPO.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/9/2026
First Seen
Valid until: 1/7/2027
Sources
Related Entities
Related Claims
UnverifiedThe fine-tuned Qwen3.5-4B model improved accuracy on BU Bench V1 from 15% to 53%.69% similarUnverifiedOn the Qwen3.5-2B model, GAD-RL achieved a Micro Recall of 59.92% on CHAOS-Bench, surpassing GRPO and fixed-weight GRPO+OPD by 8.45 and 4.43 percentage points respectively69% similarVerifiedQwen3.5's parameter activation rate is 4.3%, while DeepSeek V3's 671B model has an activation rate of 5.5%.68% similarUnverifiedOn a three-sample math problem, DeepSeek V4.1 Flash scored 74 points, on par with Opus 5.0 and surpassing GLM's 67 points and V4 Pro's 62 points.66% similarUnverifiedOn the SWE-bench Pro benchmark, the Qwen3 Coder model scored 61.7, while Opus 4.6 MAX scored 53.4.66% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/990753API
curl https://kongchang.com/api/v1/knowledge/claims/990753MCP
get_claim(id=990753)