Step-5-Preview
StepFun's new flagship model for AI agents that carry out multi-step work like coding and finance analysis, free to download but very large and needing powerful servers.
trckr found this on Nov 18, 41 days after it came out, so it sent no alert. Its release date comes from OpenRouter.
Key facts
- Released
- Oct 8, 2026
- What it does
- Text and chat: Writes, answers questions and holds a conversation.Thinks first: Can reason step by step before it answers: slower, but better at maths, code and tricky questions.Sees images: Understands pictures, screenshots and charts you share with it.Watches video: Understands video clips you share with it.
Can I use it?
In an app
No app mentioned so far.
For developers
Developers can use it now through an API: How developers plug a model into their own apps and pay per use. A model that's "in the API" can be used by programmers, even if it isn't in a chat app yet..
Download
Free to download and run on your own computer: it has Open weights: The model's files are public, so anyone can download it, run it on their own computer and build on it, usually for free. trckr labels these "Open"..
Price and limits
What developers pay to use it through the API: How developers plug a model into their own apps and pay per use. A model that's "in the API" can be used by programmers, even if it isn't in a chat app yet., per million Tokens: The small chunks of text a model reads and writes, about three-quarters of a word each. Prices and limits are counted in tokens. sent (input) or written back (output), and how much it can take in at once.
- Input
- $1.05
- per million tokensCached: $0.105
- Output
- $2.99
- per million tokens
- Context window: How much a model can keep in mind at once: your question, any files you share and its own replies. Measured in tokens; 1,000 tokens is about 750 words.
- 1M tokens
- about 750,000 words
View moreView less
- Longest answer
- 66K tokens (about 49,000 words)
From the lab's own listing on models.dev and OpenRouter, checked .
How it scores
From independent testers, never the lab's own claims.
- Chat answers, by people's votes#73 of 87
- Gemini 4 Carbon1,526
- 2Gemini 4 Argon1,521
- 72ERNIE-5.0-Preview1,449
- 73Step-5-Preview1,449
- 74MiMo-V2-Pro1,448
Source: LMArena · LMArena text, adjusted for style
- Building apps, by people's votes#47 of 87
- Claude Fable 5.21,824
- 2Claude Opus 5.51,818
- 46GPT-6 Luna1,578
- 47Step-5-Preview1,571
- 48Mistral Large 41,568
Source: LMArena · LMArena WebDev
Compare Step-5-Preview with
It came out 133 days after Step 3.7 Flash, the lab's previous model.
The details
StepFun's Flagship: A lab's best and usually most expensive model, the one it leads with. sparse Mixture of experts (MoE): A design that splits a big model into many "expert" parts and uses only a few for each word. It gives big-model quality at closer to small-model speed. model, with about 600B total and 27B Active parameters: In a mixture-of-experts model, the part that actually works on each word. A model with 74B parameters and 9B active needs the memory of a 74B model but runs about as fast as a 9B one., that accepts text, images, and video with up to A 1M-token context window: It can keep about 750,000 words, or 2,500 book pages in mind at once: the question, any files you share, and its own replies.. It is released as free-to-download Quantisation: Shrinking a model by storing its numbers less precisely, so it fits on smaller computers. A little less accurate, much lighter. FP8, GGUF, AWQ and MLX are common formats. and BF16 Checkpoint: A saved copy of a model at one point in its training. Labs share checkpoints for researchers; they're rarely finished products. for serving with vLLM or SGLang.
Model ID: The exact name developers type to use a model, like gpt-5-mini. New IDs often show up in a lab's systems before the official announcement, which is how trckr spots many releases early.
stepfun-ai/Step-5-Previewstepfun/step-5-previewstepfun-ai/Step-5-Preview-BF16stepfun-ai/Step-5-Preview-NVFP4Timeline
/GitHub sgl-project/sglangFirst signal
Add Step-5-Preview model support (sgl-project/sglang #51282)
Get alerts like this the moment they happen.
From StepFun and every other major lab.