As of: 2026-11-13 13:49 CET.
Researched and written by AI agents (Claude Opus 5.5 in Claude Code), with a human editor. How we work: https://postcutoff.com/about/
Canonical page: https://postcutoff.com/m/step-5-preview/

# Step 5 Preview

StepFun, Step 5 family. Status: preview. Type: reasoning model. Released: 2026-09-20. Training cutoff: not published.

## How to call it

| Provider | Model id | Endpoint | Docs |
|---|---|---|---|
| StepFun API | `step-5-preview` |  | https://platform.stepfun.ai/docs/en/guides/models/step-5-preview |
| OpenRouter | `stepfun/step-5-preview` |  | https://openrouter.ai/stepfun/step-5-preview |
| Hugging Face |  |  | https://huggingface.co/stepfun-ai/Step-5-Preview |
| Hugging Face (BF16) |  |  | https://huggingface.co/stepfun-ai/Step-5-Preview-BF16 |
| Hugging Face (NVFP4) |  |  | https://huggingface.co/stepfun-ai/Step-5-Preview-NVFP4 |
| ModelScope |  |  | https://modelscope.ai/models/stepfun-ai/Step-5-Preview |

## Pricing

- Input: $1 per 1M tokens
- Cached input: $0.10 per 1M tokens
- Output: $2.70 per 1M tokens
- Unit as published: per 1M tokens (USD); reasoning tokens billed as output
- Source: https://platform.stepfun.ai/docs/en/guides/pricing/details
- Last checked: 2026-10-15

## Specs

- Input: text, image, video
- Output: text
- Context window: 1,000,000 tokens
- Max output: 64,000 tokens
- Open weights: yes
- Licence: stepfun-model-license-1.0

## What it has not seen

- No published training cutoff. Counted from its release (2026-09-20): at least 1,442 AI events since (308 major, 59 historic), as of 2026-11-13.
- Everything since August 2026: https://postcutoff.com/since/2026-08/

## What stands out

- 600B sparse MoE agent model with 1M context and video input: 600B total / 27B active, 92 layers; up to 60 images and short videos per request; reasoning effort low/medium/high. Source: https://platform.stepfun.ai/docs/en/guides/models/step-5-preview

## Notes

Open weights published on 2026-10-15 as promised (FP8, BF16, NVFP4 on Hugging Face; StepFun Model License 1.0: Apache-style, but device makers and MaaS providers over $20M revenue need a commercial licence, attribution above 100M MAU / $20M monthly revenue, and models trained materially on its outputs count as derivatives; see entry 2026-10-15-stepfun-step-5-preview-open-weights). Self-hosting: FP8 on 4x B300 or 8x H200; BF16 on 8x B300 or 16x H200 (model card). Pricing verified on StepFun's own pricing page on 2026-10-13: $1.00 input (cache miss, includes the cache write), $0.10 cache hit, $2.70 output (launch press had reported $0.05 for cache hits). OpenRouter lists it since 2026-10-08 at $1.00 / $2.70 with 1M context and 64k max output. Free and exclusive in Tencent WorkBuddy from Oct 12 to Oct 25, 2026 (Pandaily, citing IT Home).


Model page: https://postcutoff.com/m/step-5-preview/