Perplexity's open-weight decision model that picks among given options for routing and checks in AI systems
Step 5 Preview
New· 4A 600 billion parameter mixture-of-experts model for agentic work, with a documented one million token window
About Step 5 Preview
On 20 September 2026 StepFun announced Step 5 Preview, its flagship model for agentic work, and the API and the company's studio went live that day. Its design is a sparse mixture of experts: 600 billion parameters overall, with 27 billion engaged for each token. StepFun documents a one-million-token context window, and the model accepts images and video as well as text, with text as its output.
What the vendor charges
StepFun's pricing table lists separate rates for input on a cache miss, input on a cache hit (a 95% discount) and output. A note beneath it explains that output tokens also cover the reasoning the model performs before replying, a cost that buyers of reasoning models often underestimate.
An outside measurement
The figure worth having does not come from the vendor. Artificial Analysis gives the model a 44 on its Intelligence Index, which puts it 24th among 200 tracked models. On StepFun's own API it measured roughly 99.8 output tokens per second, with a wait of 2.96 seconds for the first token. It also found the model wordy: it produced 160 million output tokens during the index run, and output is the costly side of the invoice.
Two caveats
The open weights are a promise and not yet a fact. StepFun's materials give 15 October 2026 as the date for a BF16 release. The Hugging Face repository under the model's name currently contains only a .gitattributes file, with no weights, license, model card or configuration.
The release date is also not settled. This description uses 20 September, the day StepFun announced the model and opened the API, while Artificial Analysis dates it 18 September, when it ran its measurements. Third-party configurations listing a 350,000-token window instead of the announced one million have been reported, and the vendor's documentation, which states one million, does not resolve that.
TrustList summary
Step 5 Preview is the flagship model for agentic work from StepFun. StepFun announced it on 20 September 2026 and opened its API and studio the same day. The architecture is sparse, a mixture of experts with 600 billion parameters in total, of which 27 billion are active on each token. Its documented context window is one million tokens.
Features
- Image and video input: accepts images and video natively, with text output.
- Long context: documented as a one million token window.
- Measured speed: Artificial Analysis clocks it at 99.8 output tokens a second, with 2.96 seconds before the first token, on StepFun's own API.
- Measured intelligence: Artificial Analysis places it 44 on its Intelligence Index, 24th of the 200 models it tracks.
Who it serves
Developers building agentic applications that need a long context window and image or video input.
Key facts
- Verbosity: Artificial Analysis counts 160M output tokens across its index, which matters because output is the costlier side of the bill.
- Open weights: StepFun's materials give a BF16 release date of 15 October 2026, but the repository held no weights at listing.
- Release date: Artificial Analysis dates the model 18 September, the day it measured it.
- Window: some third-party configurations list 350,000 tokens, while the vendor's documentation states one million.
Summary written by TrustList from the business's own published information, Nov 2026.
Benchmarks & AI stats
| Benchmark | Official | Community avg |
|---|---|---|
| Artificial Analysis Intelligence Index v4.3.2llm | 44 index | - |
| Artificial Analysis Intelligence Index v5.0.1llm | 29 index | - |
“Official” values are editor-approved and feed the ranking. “Community avg” is the mean of member submissions (shown for transparency; it never affects the ranking until an editor approves a value).
Sign in to add a benchmark score for this model.
Step 5 Preview launches
Trust Score
An earned signal from verification, reviews, awards, transparency and engagement - the vendor can't buy it.
Updated 9/20/2026
Request a demo or quote from Step 5 Preview
Reviews
Write the first review of Step 5 Preview
Used it? Your experience helps other buyers decide.
Write a reviewQuestions & answers
No questions yet. Be the first to ask about Step 5 Preview.
Step 5 Preview alternatives
Similar listings buyers compare against this one.
Microsoft AI's text-to-speech model for keeping one voice across 23 languages, with voice cloning from short audio
Anthropic's smallest and fastest Claude 5.5 model, available on its own API and three cloud platforms
Open-source 2B decision model from AWS's Strands Agents project that picks options and returns scores with a reliability figure
Google's open multimodal embedding model that maps text, images, audio and video into one vector space, under Apache 2.0
Open-weight translation model from Bilibili's Index team, covering 150 languages and following glossary and style instructions
Aleph Alpha's open-weight English-German mixture-of-experts model for self-hosting, released under Apache 2.0 in October 2026
JetBrains' open 12B mixture-of-experts coding model with 2.5B parameters active per token under Apache 2.0
More on TrustList
Everything here links back to the same verified catalogue. Pick your next stop.
- Write a reviewHelp other buyers decide
- CompaniesAgencies, consultancies and IT service providers, ranked by verified reviews.
- ProductsSoftware and SaaS with pricing, features, integrations and alternatives.
- ArticlesGuides, comparisons and research from the editorial desk and the community.
- AwardsAnnual recognition decided by verified reviews and an independent jury.
- LaunchesNew products and releases, voted up by the community every day.
- RequestsBuyers describe what they need; vendors respond directly.
- PeopleReviewers, authors and makers with public profiles.
- ComparePut up to four listings side by side before you shortlist.