Meta Releases Muse Spark 1.2 with Coding Enhancements
Meta launched Muse Spark 1.2, a coding-focused model with a 20-cent-per-million-output-tokens tier, competing on discounts.
ARTICLES BY ALEX LINDGREN
LATEST NEWS
-
-
Meta Launches Muse Code AI Agent for Large Code Bases
Meta released Muse Code, an AI agent for large code bases, in beta, enabling complex software tasks across repositories.
-
Mistral's Shieldstral Model Matches Larger Safety Models
Mistral's 3-billion-parameter Shieldstral model achieves an F1 score of 84.9% on text safety benchmarks, matching a seven-times-larger model.
-
Google to Discontinue Google Assistant in 2026 as Gemini Takes Over
Google will shut down Google Assistant on Android and Wear OS starting September 4, 2026, replacing it with Gemini, according to 9to5Google.
-
Black Forest Labs Launches FLUX 3 Video
Black Forest Labs has released FLUX 3 Video, a model that generates HD and Full HD clips up to 20 seconds long.
-
Google to retire Assistant on phones by September 4
Google will shut down its mobile Assistant service on September 4, forcing users to switch to Gemini. The change affects phones, smartwatches, and Android Auto.
-
Hark Launches Browser Agent Hark Handoff for Task Automation
Hark, a $700 million-funded startup, unveiled its browser agent Hark Handoff, which can complete tasks like ordering food or booking travel tickets.
-
OpenAI Launches New Education Plugins for ChatGPT Work and Codex
OpenAI announced three new education plugins for ChatGPT Work and Codex, designed to help K–12 educators, college educators, and students leverage AI in learning and teaching.
-
Mistral Releases Shieldstral, 3B Multimodal Safety Classifier
Mistral announced Shieldstral, a 3B open-weights multimodal safety classifier that outperforms models up to 7x its size on text safety benchmarks. Released under Apache 2.0, it runs efficiently on a single 16GB GPU.
-
Google Announces Three New Gemini Models for AI Agents
Google released three new Gemini models in July 2026 to help developers build scalable AI agents and improve robotics capabilities.
-
Bekko Embedding: Tiny Multilingual Retrieval Models
Hugging Face releases two tiny multilingual retrieval models, bekko-a8m and bekko-a25m, with 7.67M and 24.93M active parameters, outperforming larger models on benchmark tasks.
-
Alibaba's Qwen Model Promotes AI as Job Enhancer
Alibaba's Qwen 3.8 model shifts focus from job loss to productivity, aligning with market trends in 2026.