NEWSAR
Multi-perspective news intelligence
SRCSouth China Morning Post
LANGEN
LEANCenter-Right
WORDS226
ENT8
TUE · 2026-02-03 · 09:30 GMTBRIEF NSR-2026-0203-12915
News/‘Punches above its weight’: compact AI model from China’s St…
NSR-2026-0203-12915News Report·EN·Technology

‘Punches above its weight’: compact AI model from China’s StepFun outshines larger rivals

Chinese AI start-up StepFun has launched Step 3.5 Flash, a compact AI model designed for advanced reasoning and efficiency. Despite having only 196 billion parameters, significantly fewer than competitors like Moonshot AI and DeepSeek, StepFun claims its model outperforms these larger systems in benchmark tests measuring agentic, reasoning, and coding capabilities.

Ben JiangSouth China Morning PostFiled 2026-02-03 · 09:30 GMTLean · Center-RightRead · 1 min
‘Punches above its weight’: compact AI model from China’s StepFun outshines larger rivals
South China Morning PostFIG 01
Reading time
1min
Word count
226words
Sources cited
1cited
Entities identified
8entities
Quality score
100%
§ 01

Briefing Summary

AI-generated
NEWSAR · AI

Chinese AI start-up StepFun has launched Step 3.5 Flash, a compact AI model designed for advanced reasoning and efficiency. Despite having only 196 billion parameters, significantly fewer than competitors like Moonshot AI and DeepSeek, StepFun claims its model outperforms these larger systems in benchmark tests measuring agentic, reasoning, and coding capabilities. Step 3.5 Flash reportedly topped four reasoning benchmarks, surpassing leading systems from DeepSeek, Moonshot AI, Zhipu AI, and MiniMax, while trailing only OpenAI in some tests. StepFun's CTO emphasized the model's focus on strong logic, efficient context window, and speed, specifically for use in AI agents. The Shanghai-based company aims to demonstrate that smaller, more focused models can rival the performance of larger, more resource-intensive AI systems.

Confidence 0.90Sources 1Claims 5Entities 8
§ 02

Article analysis

Model · rule-based
Framing
Technology
Economic Impact
Tone
Measured
AI-assessed
CalmNeutralAlarmist
Factuality
0.70 / 1.00
Factual
LowHigh
Sources cited
1
Limited
FewMany
§ 03

Key claims

5 extracted
01

DeepSeek V3.2 has 671 billion parameters.

statisticarticle
Confidence
1.00
02

Moonshot AI’s Kimi K2.5 has 1 trillion parameters.

statisticarticle
Confidence
1.00
03

Step 3.5 Flash has approximately 196 billion parameters.

statisticarticle
Confidence
1.00
04

StepFun has unveiled a lightweight AI model called Step 3.5 Flash.

factualarticle
Confidence
1.00
05

Step 3.5 Flash outperformed larger rivals across several benchmark tests.

factualStepFun
Confidence
0.80
§ 04

Full report

1 min read · 226 words
Chinese artificial intelligence start-up StepFun has unveiled a lightweight AI model that it says punches above its weight, rivalling larger systems from domestic competitors including DeepSeek and Moonshot AI as competition intensifies in the country’s AI sector.The Shanghai-based AI lab said on Monday its latest Step 3.5 Flash model was designed to deliver advanced reasoning and agentic capabilities while maintaining efficiency.Despite its relatively modest size of about 196 billion parameters – far smaller than Moonshot AI’s Kimi K2.5 with 1 trillion parameters or DeepSeek V3.2 with 671 billion parameters – Step 3.5 Flash outperformed its larger rivals across several benchmark tests measuring agentic, reasoning and coding capabilities, according to the company’s self-reported results.Parameters are the variables that encode an AI system’s “intelligence”, with a larger number usually indicating stronger performance.Step 3.5 Flash topped four reasoning benchmarks, including AIME 2025 and IMOAnswerBench, outperforming leading systems from DeepSeek, Moonshot AI, Zhipu AI and MiniMax, and trailing only Microsoft-backed OpenAI in certain tests.StepFun says its lightweight model was designed to deliver advanced reasoning and agentic capabilities while maintaining efficiency. Photo: HandoutThe model’s compact size and focus on reasoning were deliberate choices, according to Zhu Yibo, StepFun’s co-founder and chief technology officer. Zhu said the team had prioritised “strong logic capability, efficient context window and fast speed” when developing the new system, which was purpose-built for the AI agent era.
§ 05

Entities

8 identified
§ 06

Keywords & salience

10 terms
artificial intelligence
1.00
ai model
0.90
stepfun
0.80
reasoning capabilities
0.70
agentic capabilities
0.70
lightweight model
0.60
parameters
0.50
benchmark tests
0.50
deepseek
0.40
moonshot ai
0.40
§ 07

Topic connections

Interactive graph
Network visualization showing 8 related topics
View Full Graph
Person Organization Location Event|Click node to navigate|Edge numbers = shared articles