跳转到主要内容
AITier

Seed 2.1 Turbo

ByteDanceProprietary

描述

ByteDance's low-cost, low-latency next-generation agent model for large-scale production. A deep-thinking model with the full feature set of Seed 2.1 Pro and comparable performance, it is designed for enterprise-grade deployments handling high volumes of online calls across coding, long-chain agents, and multimodal tasks. Served via Volcano Engine as Doubao-Seed-2.1-turbo.

发布日期
2026-06-24
参数规模
—
上下文长度
262K
支持模态
image, text, video

能力雷达图

60
general
50
coding
60
reasoning
51
science估算
70
agents
70
multimodal

缺少专门科学评测时，Science 由 LLM Stats 科学得分或推理能力估算。

排行榜排名

领域#排名分数来源
智能体能力模型榜48
49.0
LS
多模态榜15
65.0
LS

基准测试分数 (LLM Stats)

(LLM Stats (zeroeval))

Agents

Web Bench73.6%自报
OfficeQA Pro71.1%自报
MobileWorld70.0%自报
OneMillion Bench66.6%自报
SeedClawBench63.8%自报
WildClawBench62.8%自报
xDailyBench56.4%自报
Finance Agent v1.156.0%自报
Workspace Bench54.7%自报
Agent Startup Bench54.0%自报
Program Bench49.4%自报
Toolathlon49.1%自报
Doubao Multi-Turn Bench49.0%自报
PresentBench48.3%自报
ClawEval-MM46.0%自报
CreativeWork34.5%自报
SWE-Atlas30.6%自报
GameWorld25.9%自报
PostTrainBench18.3%自报

Chemistry

SuperChem56.6%自报

Code

CyberGym67.0%自报
Artifacts Bench47.0%自报
NL2Repo43.7%自报
DeepSWE23.0%自报

Coding

AetherCode67.9%自报
Trae Code Gen59.7%自报
Trae Error Fix56.7%自报
Repo Env46.7%自报

General

GDPval82.7%自报

Knowledge

KINA46.6%自报
MSQA42.0%自报

Math

MathVision92.7%自报
MathVista90.5%自报
MathVerse89.2%自报
Beyond AIME88.0%自报
DynaMath68.1%自报
MathArena Apex35.4%自报
LiveMathematicianBench27.7%自报
HorizonMath2.0%自报

Multimodal

Video-MME89.0%自报
OSWorld76.4%自报

Reasoning

CharXiv-D94.6%自报
BrowseCompOpenAI (2025)84.9%自报
CharXiv-R83.6%自报
MCP Atlas80.3%自报
FrontierScience Olympiad76.0%自报
Terminal-Bench 2.167.6%自报
SuperGPQA67.4%自报
ArcAGI261.3%自报
SciCode57.8%自报
SWE-Bench ProPrinceton NLP (2024)57.0%自报
Humanity's Last Exam54.6%自报
FrontierCS50.8%自报
FrontierScience Research33.3%自报
APEX-Agents29.2%自报

Video

OVOBench79.2%自报
LiveSports-3K77.1%自报
VideoSimpleQA71.4%自报
OVBench69.7%自报
VideoHolmes67.6%自报
Minerva65.9%自报
CrossVid63.2%自报
ContPhy61.1%自报
TOMATO56.8%自报

Vision

RealWorldQA86.3%自报
DUDE83.1%自报
EmbSpatialBench0.82 / 100自报
MMMU-Pro82.2%自报
LongVideoBench80.6%自报
BLINK79.4%自报
EMMA78.4%自报
TVBench77.2%自报
MMLongBench-128K76.9%自报
LVBench76.8%自报
MotionBench74.8%自报
ERQA71.3%自报
TreeBench71.1%自报
SimpleVQA0.71 / 100自报
ChartQAPro70.9%自报
VLMsAreBiased68.3%自报
WorldBench63.7%自报
BabyVision62.9%自报
OCRBench_V262.8%自报
MeasureBench58.9%自报
VisuLogic0.53 / 100自报
WorldVQA48.6%自报
VisFactor43.9%自报
Image2FloorPlan35.9%自报
MMSIBench31.4%自报
ZEROBench0.11 / 100自报

AA 评测指数

(Artificial Analysis)

暂无 AA 评测数据

LLM Stats 分类评分

(LLM Stats (zeroeval))
Structured Output
90
Legal
80
Long Context
80
Search
80
Spatial Reasoning
80
Embodied
80
3d
80
Image To Text
70
Multimodal
70
Safety
70
Finance
70
Healthcare
70
Coding
70
Economics
70
Tool Calling
70
Video
70
Vision
70
Knowledge
60
Math
60
Physics
60
Reasoning
60
General
60
Agents
60
Biology
60
Chemistry
60
Science
50
Frontend Development
50
Code
50
Systems
20

定价

输入价格$0.5 / 1M tokens
输出价格$2.5 / 1M tokens
混合价格(3:1)$1 / 1M tokens

速度

暂无速度数据

供应商价格排行

供应商价格排行

4 个供应商

最便宜: ByteDance最贵: Kilo Gateway
供应商输入输出
1ByteDance主要
$0.5
$2.5
2NanoGPT
$0.5
$2.5
3OpenRouter
$0.5
$2.5
4Kilo Gateway
$0.5
$2.5

比较该模型在不同 API 供应商之间的定价。

外部链接

LLM StatsArtificial Analysis