# SambaNova vs StepFun

> SambaNova is a hardware-driven host for fast decode on big open models. StepFun is a Shanghai lab making cheap multimodal Step models, many Apache 2.0. Host speed versus model economics.

Canonical: https://www.subconscious.dev/compare/sambanova-vs-stepfun · By The Subconscious Team · Updated September 30, 2026

## How they compare

StepFun wins on cost through model design. Step 3.7 Flash is a 198B mixture-of-experts vision-language model with only 11B active, 256K context and Apache 2.0 weights, priced at $0.20 in and $1.15 out on StepFun's API. The lab's research on attention and disaggregation aims to keep decoding cheap even on low-end accelerators. SambaNova wins on speed through hardware. Its RDU chip and three-tier memory host large models like MiniMax M2.7 and DeepSeek, with decode speeds reported by Artificial Analysis. StepFun is a lab that also serves its models; SambaNova serves other labs' models.

Choose StepFun for image and video understanding in cost-sensitive agents, or to self-host a model with few active parameters. Its first-party inference is China-hosted and Western distribution is thin, so many teams would run its open weights elsewhere. Choose SambaNova for interactive coding agents on large text models where tokens per second shape the experience. StepFun trails frontier models on hard multimodal reasoning, and SambaNova's largest claims rest on SN50 hardware still ramping.

## What each one does

### SambaNova

SambaNova designs its own inference chip, the Reconfigurable Dataflow Unit, and sells fast tokens on large open models through SambaCloud. The RDU maps the model graph onto the chip to cut trips to off-chip memory. A three-tier memory design of SRAM, HBM and bulk DRAM lets one system host very large models and hot swap between several of them in milliseconds. SambaCloud serves models like MiniMax M2.7, DeepSeek, Gemma 4 31B and GPT-OSS 120B, with speeds reported by Artificial Analysis.

### StepFun

StepFun is a Shanghai AI lab known for efficient multimodal models, with a mix of proprietary API models and open-weight releases. Its current workhorse, Step 3.7 Flash, came out in May 2026 as a 198B mixture-of-experts vision-language model with only 11B active parameters. It has 256K context, selectable reasoning levels, tool use and structured outputs, and it ships under Apache 2.0. StepFun's own API prices it at $0.20 in and $1.15 out per million tokens, and OpenRouter carries it too.

## Which is best, and when

### Choose SambaNova for

- Fast decode on large open text models.
- Copilots where generation speed is the wait.
- Multi-model agents that hot swap.

### Choose StepFun for

- Cheap vision and video understanding.
- Self-hosting Apache 2.0 weights with 11B active.
- Multimodal agents needing 256K context on a budget.

## At a glance

| Attribute | SambaNova | StepFun |
|---|---|---|
| Model access | Open weights | Open (Apache 2.0) and API models |
| Flagship models | MiniMax M2.7, GPT-OSS 120B, DeepSeek | Step 3.7 Flash, Step3 |
| Speed | ~820 tok/s on MiniMax M2.7 (SN50) | ~128 tok/s on Step 3.7 Flash |
| Price | $0.22 in, $0.59 out (GPT-OSS 120B) | $0.20 in, $1.15 out (Step 3.7 Flash) |
| Customization | - | Open weights to fine-tune |
| Deployment | SambaCloud, racks for neoclouds | First-party API, OpenRouter |
| Long context | Up to 192K (MiniMax M2.7) | 256K |

## FAQ

### What is the difference between SambaNova and StepFun?

SambaNova is a hardware-driven host for fast decode on big open models. StepFun is a Shanghai lab making cheap multimodal Step models, many Apache 2.0. Host speed versus model economics.

### When should I choose SambaNova over StepFun?

Fast decode on large open text models; Copilots where generation speed is the wait; Multi-model agents that hot swap.

### When should I choose StepFun over SambaNova?

Cheap vision and video understanding; Self-hosting Apache 2.0 weights with 11B active; Multimodal agents needing 256K context on a budget.

### Is SambaNova or StepFun cheaper?

SambaNova: $0.22 in, $0.59 out (GPT-OSS 120B). StepFun: $0.20 in, $1.15 out (Step 3.7 Flash). The cheaper choice depends on the model and workload.

### Which has more context, SambaNova or StepFun?

SambaNova: Up to 192K (MiniMax M2.7). StepFun: 256K.

## Running long-horizon agents?

If your agents run past 200K tokens, compare both against Subconscious: [Subconscious vs SambaNova](https://www.subconscious.dev/compare/subconscious-vs-sambanova.md), [Subconscious vs StepFun](https://www.subconscious.dev/compare/subconscious-vs-stepfun.md).

Full profiles: [SambaNova](https://www.subconscious.dev/providers/sambanova.md), [StepFun](https://www.subconscious.dev/providers/stepfun.md).