Seed2.1 Pro

Seed2.1 Pro ranks first on 18 of the 75 benchmarks with five or more models scored.

ByteDance Seed Released June 23rd, 2026 Family Seed 2 The lab's model card Compare with others

101

Benchmarks with a score

Ranked first, out of five or more models

In the top three

ByteDance Seed Jun 2026

Seed2.1 Pro

Closed weights Flagship

154.6 #63 of 404

Self-Reported Index, from the lab's own numbers. No Epoch Capabilities Index score yet.

MMMU-Pro 82.7% #5 /90
SciCode 59.8% #1 /49
NL2Repo-Bench 47% #17 /30

Where it ranks

Its best score on each benchmark, and where that ranks among every model with a score there. We rank each model once, at its best setting.

Knowledge and language

Benchmark	Score	Rank	Source
SuperGPQA	70.8%	#5 of 35	Lab model card
Humanity's Last Exam	55.7%	#23 of 154	Lab model card
SimpleVQA	74.5%	#5 of 30	Lab model card
MMLongBench	78.3%	#1 of 4	Lab model card
Video SimpleQA	76.4%	#1 of 4	Lab model card
HLE-Verified	42.9%	#4 of 4	Lab model card
Doubao Multi-Turn Bench	52.5%	#1 of 1	Lab model card
KINA	48.3%	#1 of 1	Lab model card
MSQA	50.2%	#1 of 1	Lab model card
xDailyBench	61%	#1 of 1	Lab model card

Math and science

Benchmark	Score	Rank	Source
Beyond AIME	87%	#1 of 14	Lab model card
FrontierScience	75%	#3 of 8	Lab model card
ARC-AGI-2	62.5%	#12 of 24	Lab model card
FrontierScience-Olympiad	75%	#1 of 2	Lab model card
xbench-ScienceQA	71%	#15 of 29	Official leaderboard
MathArena (Apex, IMO, USAMO)	31.3%	#9 of 12	Lab model card
FrontierScience-Research	28.3%	#4 of 5	Lab model card
IMO-ProofBench	54.3%	#4 of 5	Lab model card
IMO 2025	65.2%	#5 of 6	Lab model card
ExplorationBench	40%	#9 of 10	Official leaderboard
HorizonMath	2.00%	#3 of 3	Lab model card
IPhO 2025	79.3%	#1 of 1	Lab model card
LiveMathematicianBench	20.9%	#1 of 1	Lab model card

Coding

Benchmark	Score	Rank	Source
SciCode	59.8%	#1 of 49	Lab model card
AetherCode	65.8%	#1 of 3	Lab model card
Code Arena	1,519	#55 of 128	Official leaderboard
SWE-Bench Pro	57.5%	#38 of 81	Lab model card
Monthly-SWEBench	53%	#6 of 12	Official leaderboard
NL2Repo-Bench	47%	#17 of 30	Lab model card
Image-to-WebDev Arena	1,505	#38 of 55	Official leaderboard
EvoCode-Bench v2	6.70%	#11 of 13	Official leaderboard
SWE Atlas	35.2%	#13 of 14	Lab model card
DeepSWE	32.7%	#54 of 56	Lab model card
ProgramBench tests passed	50.3%	#10 of 10	Lab model card
PostTrainBench	16.5%	#7 of 7	Lab model card

Agents and tools

Benchmark	Score	Rank	Source
OSWorld	78.8%	#1 of 13	Lab model card
MCP Atlas	83.8%	#8 of 46	Lab model card
SeedClawBench	66.6%	#1 of 5	Official leaderboard
BrowseComp	86.2%	#18 of 81	Lab model card
WeaveBench	30.7%	#3 of 9	Official leaderboard
SkillsBench	60.4%	#5 of 14	Lab model card
Toolathlon	50.6%	#11 of 29	Lab model card
WildClawBench	61.7%	#2 of 5	Lab model card
Terminal-Bench 2.1	71%	#37 of 56	Lab model card
MobileWorld	73.1%	#2 of 3	Lab model card
Long-Horizon-Terminal-Bench 2.0	46.4%	#11 of 15	Official leaderboard
Long-Horizon-Terminal-Bench (LHTB)	28.6%	#19 of 25	Official leaderboard
CreativeWork	42.5%	#1 of 1	Lab model card
GameWorld	31.2%	#1 of 1	Lab model card

Professional work

Benchmark	Score	Rank	Source
GDPval	87.9%	#1 of 10	Lab model card
OfficeQA Pro	70.9%	#3 of 16	Lab model card
Agent Startup Bench	68.8%	#1 of 5	Official leaderboard
APEX-Agents	33.8%	#6 of 12	Lab model card
Vals Finance Agent	60.7%	#5 of 10	Lab model card
$OneMillion-Bench	23.7%	#8 of 13	Official leaderboard
BlenderWorld	51.3	#9 of 13	Official leaderboard
Agents' Last Exam	19.5%	#26 of 26	Lab model card
PresentBench	54.6%	#3 of 3	Lab model card
WorkSpaceBench	53%	#3 of 3	Lab model card

Vision, audio and robotics

Benchmark	Score	Rank	Source
MathVista	90.7%	#1 of 70	Lab model card
BLINK	81.4%	#1 of 29	Lab model card
MMMU-Pro	82.7%	#5 of 90	Lab model card
Video-MME	89.2%	#4 of 51	Lab model card
MathVerse	89.7%	#1 of 12	Lab model card
LongVideoBench	80.6%	#1 of 11	Lab model card
MATH-Vision (MathVision)	94.5%	#6 of 51	Lab model card
RealWorldQA	86.7%	#5 of 38	Lab model card
ERQA	72%	#5 of 36	Lab model card
CharXiv (DQ)	95.5%	#1 of 7	Lab model card
TreeBench	71.1%	#1 of 7	Lab model card
VisuLogic	54.3%	#1 of 7	Lab model card
MINERVA	70.7%	#1 of 6	Lab model card
Video-Holmes	68.2%	#1 of 6	Lab model card
LVBench	78%	#6 of 31	Lab model card
OVBench	70%	#1 of 5	Lab model card
OVO-Bench	80.7%	#1 of 5	Lab model card
TOMATO	79.5%	#1 of 5	Lab model card
CharXiv	86.4%	#16 of 75	Lab model card
MotionBench	74.9%	#2 of 9	Lab model card
DUDE	82.8%	#1 of 4	Lab model card
TVBench	80.5%	#2 of 7	Lab model card
BabyVision	73.7%	#8 of 27	Official leaderboard
CrossVid	65%	#1 of 3	Lab model card
EMMA	79.3%	#1 of 3	Lab model card
PerceptionBench (Moonshot)	55%	#6 of 16	Official leaderboard
VLMsAreBiased	83.6%	#2 of 5	Lab model card
OCRBench v2	63.2%	#9 of 22	Lab model card
WorldVQA	53%	#4 of 9	Lab model card
EmbSpatialBench	83.4%	#8 of 16	Lab model card
ChartQA Pro	70.9%	#4 of 8	Lab model card
MMSI-Bench	35.9%	#3 of 6	Lab model card
VisFactor	51.4%	#2 of 4	Lab model card
DynaMath	73.1%	#13 of 23	Lab model card
ZeroBench	22%	#20 of 31	Lab model card
ContPhy	63.6%	#2 of 3	Lab model card
LiveSports-3K	76.8%	#4 of 4	Lab model card
Image2FloorPlan	48.0	#1 of 1	Lab model card
MeasureBench	62.9%	#1 of 1	Lab model card
WorldBench	67.6%	#1 of 1	Lab model card

Safety and security

Benchmark	Score		Rank	Source
SUPERChem	59.8%		#2 of 5	Lab model card
CyberGym	70.2%		#24 of 34	Lab model card

Its lab's numbers put it at 154.6, number 63 of 404

The Self-Reported Index fits one score to the 42 capability results ByteDance Seed reported for this model, on the scale of the Epoch Capabilities Index. Labs choose which benchmarks to show and how to run them, so read it as the lab's claim, not an independent measurement. The 90% interval runs from 152.6 to 156.5. How we compute it.

Its lab's numbers next to 8 leaderboards

Where the benchmark's own leaderboard, or Epoch AI's run, also has this model. Gaps mostly come from different harnesses, effort levels and task subsets. Every pair.

Benchmark	Lab says	Board says	Gap
$OneMillion-Bench	68.8% OneMillion Bench (Table 1)	23.7% Leaderboard	+45 pts
Agent Startup Bench	68.8% Agent Startup Bench	68.8% Leaderboard	Same
BabyVision	73.7% BabyVision	73.7% Leaderboard	Same
CreativeWork	42.5% CreativeWork	42.5% Leaderboard	Same
Doubao Multi-Turn Bench	52.5% Doubao Multi-Turn Bench	52.5% Leaderboard	Same
Image2FloorPlan	48.0 Image2FloorPlan (Inhouse), Avg Score	48.0 Leaderboard	Same
SeedClawBench	66.6% SeedClawBench (blog: ClawBench)	66.6% Leaderboard	Same
xDailyBench	61% xDailyBench	61% Leaderboard	Same

Its launch post

The benchmarks the lab chose to headline, with the scores it gave.

Seed2.1 launch page and blog (Pro and Turbo) Jun 23, 2026

KINA 48.3
SuperGPQA 70.8
BeyondAIME 87
Workspace Bench (Total, 100-task OpenClaw setting) 53
Agent Startup Bench 68.8
xDailyBench 61
NL2Repo-Bench 47
ProgramBench (resolved / almost resolved / average pass rate) 0/1/50.3 50.3
Terminal Bench 2.1 71
SWE-Atlas 35.2
MathVision (w. Tool) 92.6 (94.5) 94.5
MMMU-Pro (w. Tool) 81.6 (82.7) 82.7
WorldVQA 53
ZEROBench (w. Tool) 18.0 (22.0), main set 22
BabyVision 73.7
CharXiv-RQ (w. Tool) 85.4 (86.4) 86.4
ERQA 72
MMLongBench-128K 78.3
VideoMME 89.2
TOMATO 79.5
Minerva 70.7
OVOBench 80.7
VideoSimpleQA 76.4
GDPval (model card Table 1) 87.9
Agents Last Exam (Pass@1 / Score) 19.5 / 41.4 19.5
Doubao Multi-Turn Bench 52.5
Toolathlon 50.6
SeedClawBench (blog: ClawBench) 66.6
Claw-Eval [Multimodal] (Pass^3)
Image2FloorPlan (Inhouse), Avg Score 48
MobileWorld (GUI-only) 73.1
OSWorld (Generalist CUA setting) 78.8
CreativeWork 42.5
MeasureBench 62.9
TVBench 80.5
LVBench 78
OVBench 70
SciCode 59.8
FrontierScience-Olympiad 75
MSQA 50.2

Its system card

Every result in the longer documents the lab published with it. 54 of them aren't in the launch post.

Seed2.1 Model Card Jun 23, 2026

Capabilities

Cyber risk

CyberGym 70.2

AI R&D risk

SeedKernelBench 9.21
PostTrainBench 16.5