the AI bench
VERIFIED JUNE 2026
All models

MODEL · STEPFUN · 8B (LALM)

Step-Audio 2 mini

StepFun's 8B speech-to-speech LALM trained on 8M+ hours of audio. Competitive with GPT-4o-audio on speech recognition + S2S translation benchmarks, fully open-source weights.

License: Apache 2.0 · Context: n/a · Released: August 29, 2025

The decision in five lines

The call
Buy — for voice
Best for
voice
Runs on
16 hardware picks fit (cheapest: Minisforum UM890 Pro · $463)
Watch out
Streaming real-time chat on consumer hardware — 8B LALM is heavy.
Evidence
Estimated · last verified April 2026

8B (LALM)
PARAMETERS
END-TO-END SPEECH-TO-SPEECH
TYPE
CONTEXT
~16 GB (FP16)
VRAM AT Q4

Where we recommend this

Every tier slot in the planner where this model is a top or alternate pick. Pulled live from planner.js — when the planner refreshes, this table stays current.

VOICE ·
Step-Audio 2 mini FP16 (~16 GB)StepFun 8B end-to-end speech-to-speech LALM. Competitive with GPT-4o-audio on benchmarks. Apache 2.0; full FP16 retains audio nuance.
VOICE · HIGH
Step-Audio 2 mini (8B, Apache 2.0)Unified speech-to-speech; competitive with GPT-4o-audio on several benchmarks; ~16GB FP16.

The call

StepFun's 8B speech-to-speech LALM trained on 8M+ hours of audio. Competitive with GPT-4o-audio on speech recognition + S2S translation benchmarks, fully open-source weights.

When not to use: Streaming real-time chat on consumer hardware — 8B LALM is heavy. MiniCPM-o 2.6 int4 is lighter for voice-in-voice-out on 8 GB VRAM.

Runner notes

GitHub `stepfun-ai/Step-Audio2` bundled inference scripts. No Ollama route yet. The mini-Think variant (September 2025) adds reasoning-trace output; the full 30B model promised on the StepFun roadmap was still unreleased as of late April 2026.

License
Apache 2.0
Released
August 29, 2025
Maker
StepFun

Hardware that fits

Every hardware pick whose memory fits this model at the quant we recommend. Sorted cheapest-first — the top row is your best-value fit. Click through for the full buyer’s guide.

Next step

Find-by-model — see what hardware runs this