Private AI
Step3 is a cutting-edge multimodal reasoning model—built on a Mixture-of-Experts architecture with 321B total parameters and 38B active. It is designed end-to-end to minimize decoding costs while delivering top-tier performance in vision–language reasoning. Through the co-design of Multi-Matrix Factorization Attention (MFA) and Attention-FFN Disaggregation (AFD), Step3 maintains exceptional efficiency across both flagship and low-end accelerators.
Added Jul 31, 2025
Context Window
65.5K
Max Output
8.2K
Input Price (Auto)
$0.25/1M
Output Price (Auto)
$0.65/1M
Performance metrics and benchmarks
Sourced from LMArena.
Arena Score
1348.2
Overall Rank
#174 / 367
Votes
6,541
Confidence Interval
1340.8 - 1355.6
Coding
#160 / 362
1,233 votes
1408.0
Math
#161 / 356
351 votes
1363.7
Longer Query
#177 / 345
1,259 votes
1348.0
Creative Writing
#182 / 365
785 votes
1308.1
Instruction Following
#166 / 367
1,638 votes
1345.2
Hard Prompts
#160 / 367
3,005 votes
1378.3
Auto routing is available for this model. Explicit provider selection is not available.
Loading provider options…
Spanish
#129 / 254
194 votes
1377.4
German
#131 / 275
145 votes
1361.5
Chinese
#149 / 337
233 votes
1393.6
Industry Mathematical
#149 / 350
305 votes
1378.9
Hard Prompts English
#151 / 366
1,563 votes
1402.5
Industry Business And Management And Financial Operations
#163 / 360
1,159 votes
1357.3
Industry Entertainment And Sports And Media
#165 / 365
1,101 votes
1323.2
Expert
#166 / 316
258 votes
1360.6
Industry Software And It Services
#166 / 367
2,182 votes
1391.6
English
#168 / 367
3,159 votes
1373.4
Industry Medicine And Healthcare
#171 / 337
347 votes
1365.8
Exclude Ties
#174 / 367
4,648 votes
1308.4
Korean
#178 / 245
192 votes
1251.4
Multi Turn
#178 / 365
1,113 votes
1340.8
Industry Writing And Literature And Language
#185 / 366
1,407 votes
1317.1
Russian
#185 / 330
403 votes
1326.0
Industry Life And Physical And Social Science
#191 / 365
1,079 votes
1351.8
Industry Legal And Government
#196 / 340
416 votes
1341.5
Published 2026-06-16 · Matched as step-3
LMArena Dataset