Intel / post

Fireworks serves Step 3.7 Flash, a 198B sparse MoE built for inference

Source: x.com
Date: Published Jun 4, 2026

StepFun@StepFun_ai

Great to see Step 3.7 Flash live on @FireworksAI_HQ. Designed for inference from day one, Step 3.7 Flash combines a hardware-friendly architecture with MTP-assisted decoding to reach up to 400 tokens/s. Fast, multimodal, and ready to power capable agents in real-world workflows.

Jun 4, 2026 View on X

Context

Per StepFun's post, Step 3.7 Flash is now live on Fireworks, a model-hosting platform, reaching up to roughly 400 per second through what StepFun describes as a hardware-friendly architecture combined with MTP-assisted decoding. MTP, or multi-token prediction, has the model draft additional candidate tokens ahead of the current position within the same pass, rather than generating strictly one token at a time and running the full model again for each one, cutting the number of passes needed to produce a given amount of text.

This throughput figure is consistent with StepFun's own launch specifications for the model: a 198-billion-parameter sparse model with about 11 billion active parameters per token, which StepFun's original release also reported reaching 400 tokens per second (launch specifications).

Terms in this piece · Glossary

token — The chunk of text a model reads and writes in — roughly three-quarters of a word — and the unit AI usage is billed in.
mixture-of-experts — A model built from many specialist sub-networks where only a few activate per token, giving big-model capability at small-model running cost.
LLM — A large language model — the neural network behind tools like Claude and ChatGPT, trained on huge amounts of text to predict what comes next.
AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.

Read the source x.com