Models & Releases

Rumble-2-Flash speech model launches for 40+ languages

Hugging Face: Rumble-2-Flash: an open 0.5B speech model for 40+ languages

Smaller open-source speech models could democratize voice AI for developers working with languages beyond English and major European tongues.

What happened

Amsterdam-based Reson8 released Rumble-2-Flash, a 500 million-parameter open-source speech recognition model available for free download. The model can transcribe audio in more than 40 languages and was created using on-policy distillation, where a smaller model learns from a larger production model by being corrected on its own mistakes. The model is designed for fast transcription of prerecorded audio and runs natively on standard frameworks like Transformers and vLLM.

What each outlet adds
  • Hugging Face ↗: The model achieves strong accuracy on real-world noisy phone calls while reaching over 7,000 times real-time factor on full batches.
  • SiliconANGLE ↗: In benchmark tests across seven languages, Rumble-2-Flash achieved a 7.09% average word error rate and 17.6% on lower-resource languages like Maltese and Basque, and is available free under Apache 2.0 license or through Reson8's API starting at €0.01 per hour of audio.
Why it matters

Leading speech models are trained primarily on English despite only about one in 20 people speaking it as a first language, resulting in lower accuracy for other languages. Rumble-2-Flash addresses this gap by providing developers and researchers with speech recognition built for languages beyond English and major European tongues.

Get smart on it

Rumble-2-Flash is an open source speech recognition model with 0.5 billion parameters that can transcribe audio in 40+ languages and compete with models much larger than itself. The model matters because it achieves strong accuracy on real-world noisy phone calls while running extremely fast, reaching over 7,000 times real-time factor on full batches. It was created using a technique called on-policy distillation, where a smaller student model learns from a larger production model by being corrected on its own mistakes token by token rather than learning from pre-written transcripts. The model is built on the Qwen3-ASR architecture and runs natively on standard frameworks like Transformers and vLLM without requiring custom code.

Read next
The biggest AI stories, once a day

One short email a day with the stories that matter. Free. Unsubscribe with one click.

Relevant courses

Want to go deeper than the news? Explore live, cohort-based AI courses taught by practitioners.

Browse AI courses on Maven
Related news
Models & ReleasesOpen story →

Introducing GPT-6.2 Astra and Sol

Introducing GPT‑6.2 Astra and Sol, with better professional outputs, coding, computer use, and efficiency at lower cost across ChatGPT Work, Codex, and the API.

Also reported by MarkTechPost and SiliconANGLE

Models & ReleasesOpen story →

Alibaba Releases Qwen4-Max: 2.4T MoE, 71B Active, 1M Context

Alibaba’s Qwen team has released Qwen4-Max, the first model of its Qwen4 generation. Qwen4-Max is a proprietary sparse MoE model with a 2.4T-parameter backbone and 71B parameters active per token. It accepts text, images and video, and supports a 1M-token context window. Alibaba also cut API prices and reports a 62% drop in average cost per task. TL;DR Size: 2.4T backbone parameters, 71B active per token, plus 328B N-gram Embedding parameters. 1M context, 393K max output tokens.

Also reported by SiliconANGLE

Models & ReleasesOpen story →

Microsoft rolls out new first-party models and updates to agentic AI continuous improvement

Microsoft Corp. today rolled out new generative artificial intelligence models and updates to Foundry, the company’s unified enterprise platform that helps developers and organizations ground, test and deploy AI applications and autonomous agents. Updates to Foundry IQ, part of Microsoft IQ, now provide an enterprise knowledge graph out of the box, delivering a full estate […] The post Microsoft rolls out new first-party models and updates to agentic AI continuous improvement appeared firs

Models & ReleasesOpen story →

Bland Launches Tofu-1, a Full-Duplex Speech-to-Speech Model for Calls

Bland, the San Francisco voice-AI company, has released Tofu-1, a full-duplex speech-to-speech model in research preview. The Tofu-1 speech-to-speech model targets calls from moving cars, hospital rooms and weak-signal lines. It listens while it speaks and reasons while it talks. For phone-agent teams, one model now handles audio in and audio out, with no transcript between. You can check the deployment options and specs here through our GPULeaderboard.com chrome extension. What is Tofu-1?

Models & ReleasesOpen story →

SID AI Releases SID-2 Agentic Search Model With 0.720 NDCG Swarm Mode

SID AI has released SID-2, its second-generation agentic search model, alongside SID-2 Swarm. SID-2 is an agentic search model from SID AI that searches your data, reads results and returns ranked document IDs. It replaces fixed RAG pipelines with a model that decides what to search next. TL;DR Size: Context window is 262k tokens per the report (256k on sid.ai/models). Runs on: Hosted API at platform.sid.ai, plus Baseten and SageMaker reference paths. SID’s latency figures assu

Models & ReleasesOpen story →

Cmpnd Releases dspy-wsp: Weight-Space Programming Comes to DSPy

Cmpnd has released dspy-wsp, a prototype optimizer that brings weight-space programming to DSPy. Weight-space programming is a technique in which a frontier LLM reads another model’s weights and edits them directly. There are no gradients, no training loop and no fine-tuning dataset. dspy-wsp packages this as WeightSpace, an optimizer that plugs into any DSPy program and metric. In Cmpnd’s tests, about $5 to $10 of GPT-6-Astra calls improved every decision model they tried. You can