Google's Gemini 3.5 Transcribe Handles 85 Languages and Auto-Corrects Speech
Google's new speech-to-text model, Gemini 3.5 Transcribe, supports 85 languages and reduces word error rates to 2.6 percent for recorded audio.
Google's new speech-to-text model, Gemini 3.5 Transcribe, supports 85 languages and reduces word error rates to 2.6 percent for recorded audio.
Google's Gemini Omni 1.1 Flash allows developers to extend scenes up to 40 seconds and reduce costs by up to 60% for 360p video.
OpenAI is testing a new Persistent mode for Codex, allowing AI agents to work continuously until paused, per code changes reviewed by WIRED.
Google has introduced Gemini 3.5 Transcribe, an AI model that improves voice-to-text accuracy, reducing live-speech errors to 5.5 percent.
Google's Gemini app offers multiple AI features but struggles with confusing branding, according to a recent report.
IBM announced the Granite 4.2 family, including 3B, 8B, and 30B models, trained on 15 trillion tokens.
Alibaba's Qwen3.8-Flash-Next model achieves results comparable to much larger models at one-ninth the training cost, priced at $0.16 per million input tokens.
OpenAI's CEO Sam Altman claims the company will achieve AGI by the end of 2026, according to a TIME report, with Astra serving as a key milestone.
IBM released Granite 4.2 models with 3B, 8B, and 30B parameters, offering enhanced reasoning capabilities for local deployment.
Z.ai confirmed Ox Alpha, the mysterious open-weight model, is the latest in its GLM series, with weights set to be released on Wednesday.
Perceptron, founded by former Meta scientists, launched Isaac 0.5, an open-weight vision model designed to help robots navigate complex industrial environments.
Meta plans to launch its paid AI agent Hatch in the coming weeks and release the Watermelon model in October, according to internal documents.