Our new X account is live! Follow @wizwand_team for updates

Long-context Reasoning on LongBench v2

68.2Average Score

Gemini 3 Pro

Updated 4d ago

Evaluation Results

Method
Gemini 3 Pro 2026.02	68.2	-	-	-	-	-	-	-
GLM-5 2026.02	64.5	-	-	-	-	-	-	-
Claude Opus 4.5 2026.02	64.4	-	-	-	-	-	-	-
Kimi K2.5 2026.02	61	-	-	-	-	-	-	-
DeepSeek-V3.2 2026.02	59.8	-	-	-	-	-	-	-
GPT-5.2 (xhigh) 2026.02	59.8	-	-	-	-	-	-	-
GLM-4.7 2026.02	59.1	-	-	-	-	-	-	-
Qwen3-8B-SFT w/ LongR 2026.02	36.23	38.75	32.67	39.12	41.15	33.2	-	-
Qwen3-4B-SFT w/ LongR 2026.02	35.44	38.75	37.21	26.39	35.16	35.61	-	-
Elastic Attention 2026.01	33.41	-	-	-	37.33	31.2	0.66	-
Qwen3-8B* 2026.02	33.38	35.65	28.76	34.65	40.62	28.93	-	-
Qwen3-8B-SFT w/ DAPO 2026.02	32.84	37.93	28.57	32.21	39.9	28.49	-	-
Llama-3.1-8B-Instruct 2026.01	32.69	-	-	-	32	33.08	-	-
DRIFT 2026.02	32.41	36.67	28.37	33.33	-	-	-	-
Qwen3-8B 2026.01	31.97	-	-	-	39.33	27.82	-	-
PruLong 2026.01	31.73	-	-	-	38.67	27.82	0.7	-
Qwen3-4B* 2026.02	31.46	35.16	28.1	31.46	34.7	29.46	-	-
Qwen2.5-Instruct-7B 2026.02	31.01	41.11	26.05	24.07	-	-	-	-
DuoAttention 2026.01	31.01	-	-	-	40.67	25.56	0.7	-
Elastic Attention 2026.01	30.77	-	-	-	30.67	30.83	0.75	-
Elastic Attention 2026.01	30.74	-	-	-	30.68	30.77	0.78	-
Qwen3-4B-SFT w/ DAPO 2026.02	30.41	33.33	31.36	23.87	28.32	31.63	-	-
DRIFT 2026.02	29.22	32.22	28.84	25	-	-	-	-
InfLLM-V2 2026.01	29.09	-	-	-	29.32	28.67	-	-
InfLLM-V2 2026.01	29.09	-	-	-	29.32	28.67	-	-
Elastic Attention 2026.01	29.09	-	-	-	28	29.7	0.68	-
PruLong 2026.01	28.61	-	-	-	28	28.92	0.7	-
Elastic Attention 2026.01	28.12	-	-	-	32	25.94	0.72	-
Elastic Attention 2026.01	27.88	-	-	-	34	24.44	0.7	-
DuoAttention 2026.01	27.88	-	-	-	28.67	27.44	0.7	-
xRAG 2026.02	27.43	31.11	25.58	25	-	-	-	-
COCOM 2026.02	27.24	27.78	23.26	34.26	-	-	-	-
COCOM 2026.02	27.04	27.22	23.26	34.26	-	-	-	-
COCOM 2026.02	27.04	29.44	21.4	34.26	-	-	-	-
Qwen3-8B-SFT 2026.02	26.64	32.02	24.45	23.35	28.25	25.65	-	-
DRIFT 2026.02	26.44	26.67	29.77	19.44	-	-	-	-
DRIFT 2026.02	26.24	32.22	24.65	19.44	-	-	-	-
Qwen3-4B 2026.01	25.96	-	-	-	32.67	22.18	-	-
DuoAttention 2026.01	25.96	-	-	-	28.67	24.44	0.7	-
Qwen3-4B-SFT 2026.02	25.65	32.93	22.53	19.64	21.54	28.1	-	-
InfLLM-V2 2026.01	25.48	-	-	-	28	24.06	-	-
PruLong 2026.01	24.35	-	-	-	27.6	22.33	0.7	-
NaiveRAG 2026.02	22.86	17.78	26.05	25	-	-	-	-
Mistral-7B-v0.2 2026.02	20.87	25	16.28	23.15	-	-	-	-
LLMLingua-2 2026.02	20.68	26.11	13.95	25	-	-	-	-
Qwen3-8B-Base 2026.02	20.58	27.93	17.86	14.08	19.67	21.14	-	-
Qwen3-4B-Base 2026.02	17.13	20.71	13.69	17.78	19.59	15.62	-	-
ICAE 2026.02	13.12	13.89	14.88	10.56	-	-	-	-
Base model 2026.01	-	-	-	-	-	-	-	39.4
No-Thinking 2026.01	-	-	-	-	-	-	-	30
SFT 2026.01	-	-	-	-	-	-	-	40.2
Hard-Length 16k 2026.01	-	-	-	-	-	-	-	40.9
Hard-Length 8k 2026.01	-	-	-	-	-	-	-	40.7
Hard-Length 8k → 4k 2026.01	-	-	-	-	-	-	-	38.9
Soft-Length 2026.01	-	-	-	-	-	-	-	39.4
Normalized-Length 2026.01	-	-	-	-	-	-	-	36.6
TWYN 2026.01	-	-	-	-	-	-	-	38.8
Adaptive-Answer 2026.01	-	-	-	-	-	-	-	39.4
Format-Adaptive-Answer 2026.01	-	-	-	-	-	-	-	37.4