StepFun Step 5 Preview: 600B MoE Model with 1M Context and Vision Released for Free Preview
StepFun Step 5 Preview, a new 600B-parameter, 27B-active sparse MoE model with 1M context and vision, is now free for one week on Nous Portal, matching GPT-6 Luna on the Hermes Index.
What changed?
StepFun Step 5 Preview is a new large language model released as a preview with the following features: 600B total parameters (27B active) using a sparse Mixture-of-Experts (MoE) architecture, a 1 million token context window, vision capabilities, and open weights set for release on October 15. The model scored 33.89 on the Hermes Index, which matches GPT-6 Luna. StepFun Step 5 Preview is available for free for one week via the Nous Portal.
Why does it matter to an everyday developer?
Developers have temporary free access to a high-capacity model with vision and an unusually large 1M token context window. The strong Hermes Index score means it performs at the level of GPT-6 Luna, a contemporary frontier model. Sparse MoE design offers efficiency for large-scale inference. Open weights scheduled for October 15 will allow self-hosting and experimentation beyond commercial SaaS APIs. The temporarily free access enables benchmarking, prototyping, and migration testing without initial costs.
What can the developer do now?
Developers can access StepFun Step 5 Preview through the Nous Portal at no cost for a limited time. They can evaluate its context handling, vision support, and general capabilities in comparison to their current models or products. This is an opportunity to benchmark complex workflows that require long context windows or multi-modal input. Developers interested in running models locally or on their own infrastructure should monitor the release of open weights on October 15 to plan for future self-hosted deployments.
