TII has launched Falcon-ASR, a 1.6B parameter speech-to-text model supporting Arabic and four other languages, achieving a 20.92% WER in internal tests.
A developer created a Python-based tool to identify unedited AI-generated text by combining automated pattern recognition with manual editorial standards.
The developers of the Real AI SA semantic analyzer have invested 8 ₽ million into a tool that parses text into graphs without LLMs, achieving a 40x increase in document processing speed.
Sarvam AI has introduced Saaras V4, a speech-to-text model supporting 22 Indian languages and global English, featuring a 3-billion parameter architecture and new keyterm prompting capabilities.
Supersonic Labs has released Julia 1, a 144.3M-parameter open-source decision model capable of running locally on CPUs and in web browsers.
Just AI has updated its Jay Guard security layer to improve the detection of jailbreak attempts in Russian-language AI agents.
Analysis of 12,081 messages reveals that regex-based lead generation in Telegram misses 79.5% of real inquiries while generating 60% noise, necessitating a shift to multi-stage AI pipelines.
Knowledgator has launched GLiFormer, a 575-million parameter encoder that performs nested JSON extraction without token generation, achieving high accuracy and lower latency than LLMs.
