Web Navigation on WebArena Lite

61.8Average Success Rate

Qwen3-30B-A3B

Updated 1mo ago

Evaluation Results

Method
Qwen3-30B-A3B 2025.10	61.8	-	53.3	57.1	-	-	-	71.1	-	-
Qwen3-30B-A3B 2025.10	58.2	-	53.3	54.3	-	-	-	64.4	-	-
Qwen3-4B 2025.10	57.3	-	56.7	54.3	-	-	-	60	-	-
Qwen3-4B 2025.10	54.1	-	53.3	51.4	-	-	-	56.7	-	-
Llama-3.1-8B + WebAgent-R1 2025.10	51.8	-	56.7	57.1	-	-	-	44.4	-	-
Claude 4.5 Sonnet 2025.10	50	-	50	62.9	-	-	-	40	-	-
Gemma3 + MiRA 2026.03	43	73.7	56.7	54.3	30.8	28.9	-	-	-	-
Llama3.1 + WebRL 2026.03	38.8	68.4	36.7	51.4	15.4	40	-	-	-	-
SFT collector with SRC 2026.06	36	-	-	-	-	-	-	-	31.24	11.9
Qwen3-30B-A3B 2025.10	35.5	-	43.3	40	-	-	-	26.7	-	-
Gemma3 + WebRL 2026.03	35.1	68.4	43.3	40	30.8	20	-	-	-	-
Gemma3 + DigiRL 2026.03	33.3	52.6	46.7	37.1	34.6	20	-	-	-	-
Gemini-2.5-pro-SGO 2026.03	32.1	26.3	50	37.1	23.1	28.9	-	-	-	-
Gemma3 + SFT (BC) 2026.03	30.9	52.6	40	42.9	23.1	17.8	-	-	-	-
Llama3.1 + DigiRL 2026.03	30.3	57.9	26.7	37.1	33.3	17.8	-	-	-	-
Kimi K2.5 2026.06	29	-	-	-	-	-	-	-	32.05	-
Llama3.1 + AWR 2026.03	28.5	57.9	26.7	31.4	26.7	17.8	-	-	-	-
LEAP-style post-hoc correction 2026.06	26.8	-	-	-	-	-	-	-	28.64	28.64
Qwen3-4B 2025.10	24.5	-	30	22.9	-	-	-	22.2	-	-
Expert SFT + SRC 2026.06	24	-	-	-	-	-	-	-	30.13	-
Gemini-2.5-pro 2026.03	23	31.6	30	34.3	15.4	13.3	-	-	-	-
GLM-4 + SFT (BC) 2026.03	22.4	47.4	13.3	31.4	23.3	13.3	-	-	-	-
Gemini-2.5-flash 2026.03	20.6	21.1	33.3	14.3	11.5	24.4	-	-	-	-
Llama3.1 + SFT (BC) 2026.03	20.6	36.8	6.7	20	33.3	17.8	-	-	-	-
Expert SFT 2026.06	20.5	-	-	-	-	-	-	-	15.34	-
AutoWebGLM 2026.03	18.2	9.4	15	28.6	24.8	17.1	-	-	-	-
GPT-4-Turbo 2026.03	17.6	10.5	16.7	14.3	36.7	13.3	-	-	-	-
OEC-style random switch 2026.06	17	-	-	-	-	-	-	-	16.61	10.15
Base model 2026.06	15.2	-	-	-	-	-	-	-	18.48	-
GPT-4o 2026.03	13.9	10.5	10	20	20	11.1	-	-	-	-
GLM-4-Chat 2026.03	6.1	5.3	10	6.7	3.3	6.7	-	-	-	-