Qwen3.5 9B (Non-reasoning)
AlibabaQwen開源權重Apache 2.0 · 商用許可
描述
Qwen3.5-9B is a 9 billion parameter vision-language model using Gated DeltaNet hybrid architecture with a 3:1 ratio of linear attention to full softmax attention. It supports 262K native context length and delivers strong performance across knowledge, reasoning, coding, and multilingual tasks.
發布日期
2026-03-02
參數規模
9.0B
上下文長度
262K
支援模態
image, text, video
能力雷達圖
18
general
24
coding
79
reasoning
47
science
70
agents
60
multimodal
排行榜排名
基準測試分數 (LLM Stats)
(LLM Stats (zeroeval))Agents
t2-bench
79.1%自報
BFCL-V4
66.1%自報
VITA-Bench
29.8%自報
DeepPlanning
18.0%自報
Biology
GPQANYU + Cohere + Anthropic (2023)
81.7%自報
Chemistry
SuperGPQA
58.2%自報
Communication
Multi-Challenge
54.5%自報
Finance
MMLU-Pro
82.5%自報
MMLU-ProX
76.3%自報
General
IFEvalGoogle Research (2023)
91.5%自報
MMLU-Redux
91.1%自報
C-Eval
88.2%自報
MAXIFE
83.4%自報
Global PIQA
83.2%自報
MMMLU
81.2%自報
Include
75.6%自報
LiveCodeBench v6
65.6%自報
IFBench
64.5%自報
NOVA-63
55.9%自報
LongBench v2
55.2%自報
Language
WMT24++
72.6%自報
Long Context
AA-LCR
63.0%自報
Math
HMMT 2025
83.2%自報
HMMT25
82.9%自報
PolyMATH
57.3%自報
AA 評測指數
(Artificial Analysis)Coding Index(Artificial Analysis)23.5
Intelligence Index(Artificial Analysis)20.6
Tau2(Sierra + U Toronto + Vector Institute (2025))0.9
Gpqa(NYU + Cohere + Anthropic (2023))0.8
Lcr(Artificial Analysis)0.5
Ifbench(Google Research (2023))0.4
Scicode(UIUC + Argonne National Lab (2024))0.3
Terminalbench V2 10.2
Terminalbench Hard(Stanford × Laude Institute (2026))0.2
Hle(Center for AI Safety + Scale AI (2025))0.1
LLM Stats 分類評分
(LLM Stats (zeroeval))Instruction Following80
Language80
Math80
Biology80
Legal70
Physics70
Reasoning70
Structured Output70
Finance70
General70
Healthcare70
Chemistry70
Tool Calling70
Long Context60
Multimodal60
Spatial Reasoning60
Economics60
Vision60
Agents50
Communication50
定價
輸入價格$0.17 / 1M tokens
輸出價格$0.25 / 1M tokens
混合價格(3:1)$0.19 / 1M tokens
速度
Tokens/秒81.9
首Token延遲0.46s
首回答延遲0.46s
供應商價格排行
供應商價格排行
1 個供應商
供應商輸入輸出
1Alibaba主要
$0.17
$0.25
比較該模型在不同 API 供應商之間的定價。