Analog Circuit Sizing on ACS In-domain 8 trained tasks (train)

0.94ACCIA ScoreTurn-Level GRPO (Ours)

Evaluation results

Method
Turn-Level GRPO (Ours)2026.01	0.94	1	0.91	1	1	0.98	0.99	0.97	0.97
Qwen3-30B-Instruct-2507 + Trajectory-Level GRPO2026.01	0.69	0.98	0.74	0.99	0.98	0.92	0.87	0.61	0.85
Qwen3-30B-Instruct-2507 + Single-Turn GRPO2026.01	0.68	0.99	0.61	0.93	0.97	0.93	0.93	0.57	0.83
Bayesian Optimization2026.01	0.51	0.83	0.23	0.78	0.86	0.41	0.69	0.2	0.56
Qwen3-235B-Instruct-25072026.01	0.33	0.88	0.07	0.85	0.87	0.6	0.66	0.01	0.53
Qwen3-235B-Instruct-25072026.01	0.29	0.76	0.06	0.75	0.77	0.55	0.66	0.01	0.48
DeepSeek-V3.2 (thinking)2026.01	0.26	0.85	0.31	0.78	0.81	0.78	0.65	0.55	0.62
DeepSeek-V3.2 (thinking)2026.01	0.25	0.89	0.33	0.8	0.74	0.53	0.66	0.65	0.61
DeepSeek-V3.2 (no-think)2026.01	0.24	0.93	0.34	0.83	0.88	0.6	0.67	0.5	0.62
DeepSeek-V3.2 (no-think)2026.01	0.23	0.89	0.31	0.81	0.89	0.79	0.63	0.46	0.63
Qwen3-30B-Instruct-25072026.01	0.2	0.65	0.15	0.71	0.76	0.5	0.59	0.01	0.45
Qwen3-30B-Instruct-25072026.01	0.18	0.79	0.21	0.75	0.83	0.39	0.65	0	0.47