Meet Ultramarine 1 Flash from @paperinstr: built for professional work across finance, law, consulting, and operations.
Ultramarine 1 Flash outscores Claude Fable 5, GPT-5.6 Sol, Gemini 3.8 Flash, and DeepSeek V4.1 Flash on benchmarks including JobBench and FrontierFinance,
Qwen3-ASR 1.7B is the most accurate open-weight transcription model of its size, on the Pareto frontier of the Open ASR Leaderboard.
We optimized it to achieve the highest throughput, lowest latency, and lowest cost, at a 3,128x real-time factor on hour-long audio and $0.0013
Over the last few weeks, I've been optimizing Qwen3-ASR at Baseten. It's now the fastest, lowest-cost way to run batch transcription!
An hour of audio → transcribed in under 5 seconds.
Cost → about a tenth of a cent.
Throughput → 7x stock vLLM, on a single GPU.
The biggest
Meet NVIDIA Nemotron 3.5 Transcribe 🎙️
An open-weight, promptable 30B MoE model with ~4B active parameters that brings transcription in 40 languages, speaker attribution, translation, contextual prompting and audio understanding into one checkpoint.
Watch it keep track of who
Our recruiting team has been busy. Welcome to our newest engineering leaders: Bryan Vann, @ilina_mitra, @jaikumarganes, and Rahmi Marasli!
Bryan, Ilina, JK, and Rahmi will help scale our inference platform, work alongside customers on their hardest inference problems, and
