On September 20, 2026, StepFun — the Chinese AI lab backed by top-tier investors — officially launched Step 5 Preview with API access and Studio platform available the same day. The model features 600 billion parameters in a sparse Mixture-of-Experts (MoE) architecture, activating approximately 27B parameters per token for maximum computational efficiency. With a 1-million-token context window, multimodal input support (text and images), and advanced reasoning, Step 5 Preview scores 44 on the Artificial Analysis Intelligence Index — tying with Kimi K3 Max and Grok 4.6, and just 3 points below GPT-5.6 Sol — at a price of $1 USD per million input tokens and $2.70 USD per million output tokens.
What Did StepFun Announce with Step 5 Preview?
Step 5 Preview is a multimodal frontier-level language model built on a sparse MoE architecture (600B total parameters, ~27B activated per token), enabling high-quality inference at a dramatically reduced compute cost. Its 1-million-token context window handles full documents, large codebases, or extended conversation histories without fragmentation. At 44 points on the Artificial Analysis Intelligence Index, it ranks in the global top 5 and competes directly with the most capable Western models. Full open-source weights are confirmed for October 15, 2026. At $1/M input tokens versus $7/M for GPT-5.6 Sol, Step 5 delivers an 85% cost reduction for equivalent frontier-level capabilities.
"A frontier model at $1 per million tokens means that automating complex business processes with AI is no longer a privilege of large corporations — it's a competitive advantage within reach of any SMB."
Davarion Group & LabsReal Impact for SMBs
- 0185% cost reduction in AI inference: processing 1 million tokens (~750,000 words) costs $1 USD on Step 5 vs. $7 USD on GPT-5.6 Sol — critical for businesses processing contracts, emails, customer support, or data analysis at scale.
- 021M-token context for complete documents: analyze full contracts, entire product catalogs, complete customer histories, or whole codebases in a single API call — no chunking required that degrades coherence and accuracy.
- 03Open weights on October 15: businesses can deploy Step 5 on their own infrastructure for maximum data privacy and full control — especially relevant for healthcare, legal, and finance sectors handling sensitive information.
- 04Recommended immediate action: sign up for StepFun Studio today to benchmark Step 5 Preview against your current AI use cases, especially if you're currently paying for GPT-4o, Claude Sonnet, or Gemini Flash for high-volume text analysis or generation tasks.
The Step 5 Preview launch confirms a trend already visible with DeepSeek V4.1 and Kimi K3: Chinese AI labs are closing the performance gap with Western models while offering dramatically lower prices. For an SMB in Houston currently spending $500–$2,000/month on OpenAI credits to automate customer service, document analysis, or sales proposal generation, Step 5 represents an opportunity to cut that spend by up to 85% while maintaining frontier-level quality. The upcoming open weights release in October also opens the door to private on-premises or cloud deployments (AWS, GCP), eliminating vendor lock-in and ensuring data privacy for sensitive business operations.
At Davarion Group & Labs, we help businesses in Houston TX and across Latin America evaluate, integrate, and optimize AI models like Step 5 Preview into real workflows. From autonomous customer service agents to document analysis systems and sales automation, our team can implement these technologies in weeks — not months. If you're overpaying for AI today or haven't started your digital transformation, this launch is the perfect moment to act. Visit davarion.com to schedule a free consultation.
