Step-3
StepFun's Open Vision Frontier
StepFun's July 2025 vision-language model — 321B parameters in a mixture-of-experts arrangement, native multimodal from training rather than bolted on. Released open-weight, with later Step-3.5- Flash adding speed and Step-Audio variants extending into voice.
Intelligence
Good
Speed
Good
140 tok/s output
Cost
Free
Open weights — self-host
Context
66K
Up to 65,536 tokens
How are Intelligence, Speed & Cost bucketed?
Intelligence and Speed buckets are percentile ranks on
Artificial Analysis. Cost buckets are fixed dollar
thresholds keyed off output-token price ($/M out).
Intelligence
- Top 1%≤ 1%
- Top 5%≤ 5%
- Top 10%≤ 10%
- Good≤ 25%
- Medium≤ 50%
- Below avg> 50%
Speed
- Top 1%≥ 345 tok/s
- Top 5%≥ 237 tok/s
- Top 10%≥ 196 tok/s
- Good≥ 146 tok/s
- Medium≥ 90 tok/s
- Slow< 90 tok/s
Cost
- Freeopen weights · self-host
- Low< $1 / M out
- Moderate$1–5 / M out
- High≥ $5 / M out
Why it matters
Demonstrates that Chinese model labs are differentiating now — StepFun on multimodal, MiniMax on inference cost, Moonshot on long-horizon agents, Zhipu on coding. Not a single race anymore.
Core Capabilities
Long Documents
Handles entire codebases, books, and multi-doc RAG.
Multimodal
Combines text, vision, and audio in one model.
Generative
Produces images, video, audio, or other media.
Agent Workflows
Built for tool use and autonomous tasks.
Context Window
66k tokens
≈ 50 pages
4k Chat 聊天
32k Long docs 长文档
128k Books 整本书
400k Multi-doc 多文档
1M Codebase 整个代码库
10M
66k
Availability
API
Not available
Product / App
Not available
Open Source
Released
Enterprise
—
Pricing Model
Free / self-host
Open weights — pay only for compute.
Self-host Capability / Performance
Where this model sits relative to the middle 60% of models in the tree. All scores are 0–10 (higher is better).
Lower 20% Upper 80% This model
Context / memory
Context window size · log-scaled
6.0
9.0
5.1
Lower 20% 20th percentile — 20% of models score below this This model Where the current model lands Upper 80% 80th percentile — only 20% of models score above this
Percentile boundaries are computed across every model in the tree that reports the underlying benchmark for each capability.
What it feels like
- Vision-language model from Stepfun — see the linked sources below for benchmark and review coverage
- Tool-use and agent loops are the typical fit per the published model card
- Vision and multimodal tasks are the typical fit per the published model card
Best use cases
- Agent / tool-use workflows that match the model's published benchmarks
- Vision tasks (charts, documents, images) per the model card
- See the model spec and sources block for benchmarked use cases
Not ideal for
- Tasks far outside the modalities listed in this model's spec
- Workflows where a more recent successor in the same family scores higher