Boom. Mistral's Voxtral TTS hits, a scrappy 4B model that smokes ElevenLabs on win rates. Open weights? Yeah, but with enterprise teeth.
Forget AGI dreams—this week's AI wins are in the guts of inference. TurboQuant just gutted the KV cache bottleneck, making long contexts cheap.
70 milliseconds. That's the latency Mistral claims for its new Voxtral TTS model on a 10-second clip. Open-weight, multilingual, and gunning for ElevenLabs — but is the hype real?
Mistral was the open-source LLM rebel everyone cheered. Now they're crashing the TTS party with a model that fits on your smartwatch. Game on – or gimmick?