Sub-concept Generation on SCAR benchmark leaderboard · SOTA2 Research

Sub-concept Generation on SCAR

47Semantic Match AccuracyLlama-70B

Evaluation Results

Method	Links
Llama-70B2026.05		47
Llama-405B2026.05		47
DeepSeek-R12026.05		45
Llama-8B2026.05		44
Grok-42026.05		42
Qwen3-14B2026.05		42
GPT-4.1-mini2026.05		42
Gemini-2.52026.05		42
Qwen3-32B2026.05		40
GPT-OSS-20B2026.05		39
GPT-OSS-120B2026.05		34
GPT-4.1-nano2026.05		34
DeepSeek-R12026.05		33
Llama-405B2026.05		33
Llama-70B2026.05		32
Grok-42026.05		31
Qwen3-14B2026.05		31
GPT-4.1-mini2026.05		31
Qwen3-32B2026.05		31
Gemini-2.52026.05		29
GPT-OSS-20B2026.05		28
GPT-OSS-120B2026.05		27
Llama-8B2026.05		27
GPT-4.1-nano2026.05		27