ModelTape
All modelsNew releasesRankingsRuns on my GPUOpen the app →
ModelTape / Models / Step 5 Preview
StepFun · open-weights

Step 5 Preview

Step 5 Preview is an open-weights language model from StepFun, released September 16, 2026. It has 604B parameters (mixture-of-experts, 7B active). The context window is 1M tokens. API price: $1.05 input / $2.99 output per million tokens.

Step 5 Preview is StepFun's flagship sparse MoE model (about 600B total parameters, about 27B active per token) with text, image and video input, released via API on 18 September 2026 with open weights announced for 15 October 2026.

Check it against your own PC →

Lab
StepFun
Released
September 16, 2026
Weights
Open other
Parameters
604B MoE · 7B active
Context
1M
Price / M tokens
$1.05 / $2.99 in / out
Epoch ECI
not yet rated

Can you run Step 5 Preview locally?

Even with 32 GB of VRAM and 32 GB of RAM it is too large to run at a usable speed on a single consumer GPU. Estimates assume 32 GB of DDR5 RAM, a 16K context and llama.cpp-style GGUF quantization. Enter your exact hardware in the live app.

GPU memoryVerdictBest quantVRAMRAMtok/s
6 GB VRAMToo large––––
8 GB VRAMToo large––––
12 GB VRAMToo large––––
16 GB VRAMToo large––––
24 GB VRAMToo large––––
32 GB VRAMToo large––––

Ready-made quantizations on Hugging Face: smalinin/Step-5-Preview-GGUF · kefallrain/Step-5-Preview-EXL3-2.4bpw-2xSpark · pb09204048/Step-5-Preview-5layer

Benchmark results

Epoch AI has not published results for Step 5 Preview yet. New models are usually evaluated within days to weeks.

Links

Announcement · Hugging Face: stepfun-ai/Step-5-Preview

More from StepFun

Step 3.7 Flash · Step 3.5 Flash 2603

Data updated Thu, 12 Nov 2026 09:42 UTC. Benchmarks, Capabilities Index and model announcements: Epoch AI, “Capabilities & benchmarking” and “Data on AI models” (CC BY 4.0); category scores are ModelTape's own fit on Epoch's results. Releases and prices: models.dev (MIT). Sizes and quantizations: Hugging Face. “Runs on” figures are estimates from model size, quantization and memory bandwidth (±30–50%).
Live app · All models · New releases · Rankings · 6 GB · 8 GB · 12 GB · 16 GB · 24 GB · 32 GB · Legal notice (Impressum) · Privacy