THE BIG ONE
Liquid AI Launches RIPL, a Real-Time Inference Engine — Liquid AI has rolled out RIPL, a real-time inference engine for its LFM models, boasting impressive latency improvements. On one B200, the LFM2.5-1.2B model on RIPL outperformed Qwen3.5-0.8B with p95 latency reductions between 14.4 to 28.4 times. This means developers can now leverage faster response times in their applications, making it a game-changer for real-time systems and interactive AI. Read more →
QUICK HITS
Gradium Open-Sources Phonon, a 31M-Parameter On-Device TTS Model — Gradium has released Phonon, a compact and efficient TTS model that fits within 64.5 MB and delivers speech in just 7 ms. Why it matters: This model allows developers to integrate high-quality speech synthesis directly on devices, enhancing user experience without relying on cloud services. Read more →
Cohere Releases Rerank 5 with Passage-Level Context Focus — Cohere’s Rerank 5 introduces a 32K context window and improved relevance scoring, enhancing search capabilities for Pro and Fast users. Why it matters: This upgrade enables more accurate retrieval of information, crucial for applications like customer support chatbots and research tools. Read more →
IBM AI Releases Granite Speech 5.0 2B for Multilingual Tasks — The new Granite Speech model excels in transcription and translation, ranking #2 on the Open ASR Leaderboard. Why it matters: This model supports global communication needs, making it easier for businesses to engage with diverse markets. Read more →
Recursive Releases Flint, a Local Coding Model — Flint, a new coding model with 40B total parameters, is now available for local use, offering impressive performance on coding tasks. Why it matters: This allows developers to run powerful language models on local machines, improving privacy and reducing dependency on cloud resources. Read more →
ONE THING TO TRY
Tinker with Gradium's Phonon model for your next project. Set it up on your local machine to explore on-device text-to-speech capabilities and enhance user interactions.
SIGN-OFF
That's it for this week! Keep experimenting with these tools, and let me know what you think!