Qwen3.5 0.8B (Non-reasoning)
AlibabaQwen開源權重Apache 2.0 · 商用許可
描述
Qwen3.5-0.8B is a 0.8 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 features both thinking and non-thinking modes.
發布日期
2026-03-02
參數規模
800M
上下文長度
—
支援模態
—
能力雷達圖
5
general
1
coding
24
reasoning
18
science
20
agents
10
multimodal
排行榜排名
基準測試分數 (LLM Stats)
(LLM Stats (zeroeval))Agents
t2-bench
11.6%自報
Chat
IFEvalGoogle Research (2023)
44.0%自報
Multi-Challenge
18.9%自報
General
C-Eval
50.5%自報
NOVA-63
42.4%自報
Include
40.6%自報
MAXIFE
39.2%自報
Instruction Following
IFBench
21.0%自報
Language
MMLU-Redux
59.5%自報
MMMLU
44.3%自報
MMLU-Pro
42.3%自報
MMLU-ProX
34.6%自報
WMT24++
27.2%自報
Long Context
LongBench v2
26.1%自報
Math
PolyMATH
8.2%自報
Reasoning
Global PIQA
59.4%自報
SuperGPQA
21.3%自報
GPQANYU + Cohere + Anthropic (2023)
11.9%自報
AA-LCR
4.7%自報
Tool Calling
BFCL-V4
25.3%自報
AA 評測指數
(Artificial Analysis)Tau2(Sierra + U Toronto + Vector Institute (2025))65.2
Gpqa(NYU + Cohere + Anthropic (2023))23.6
Ifbench(Google Research (2023))21.6
Lcr(Artificial Analysis)8.0
Intelligence Index(Artificial Analysis)5.4
Hle(Center for AI Safety + Scale AI (2025))5.1
Coding Index(Artificial Analysis)1.2
Terminalbench V2 10.4
Terminalbench Hard(Stanford × Laude Institute (2026))0.0
LLM Stats 分類評分
(LLM Stats (zeroeval))Language40
Math40
Structured Output40
Chat30
Instruction Following30
Legal30
Physics30
Reasoning30
Finance30
General30
Healthcare30
Long Context20
Agents20
Chemistry20
Communication20
Economics20
Tool Calling20
Multimodal10
Spatial Reasoning10
Biology10
Vision10
定價
輸入價格免費
輸出價格免費
混合價格(3:1)免費
速度
Tokens/秒0.0
首Token延遲0.00s
首回答延遲0.00s
供應商價格排行
暫無提供商資料