Amazon Bedrock adds reinforcement fine-tuning with OpenAI-compatible APIs

The Allen Institute for AI has released MolmoWeb, a fully open web agent that operates websites using only screenshots without access to source code. The 8B-parameter model achieves 78.2% success on WebVoyager—nearly matching OpenAI's o3 at 79.3%—while being trained on one of the largest public web task datasets ever released.

Google released Lyria 3 Pro, an updated music generation model capable of creating full 3-minute songs—six times longer than the 30-second limit of its predecessor launched last month. The new version adds granular control over song structure, allowing users to specify intros, verses, choruses, and bridges. It's available now for paid Gemini users, enterprise customers, and developers via API.

OpenAI announced the shutdown of its Sora video generation app on Tuesday via an X post, just two days after publishing usage guidelines and following Disney's withdrawal from a proposed $1 billion investment deal. The move represents OpenAI's second major product discontinuation in recent months, after deprecating GPT-4o in January with two weeks' notice.

Google announced Lyria 3 Pro, expanding the Gemini app's music generation capability from 30-second tracks to full 3-minute songs. The model improves structural understanding of musical composition, allowing users to prompt for specific elements like intros, verses, choruses, and bridges. Available now for Gemini subscribers with tier-based daily limits (10-50 tracks/day) and in Vertex AI, Google AI Studio, and the Gemini API for developers.

Google announced Lyria 3 Pro, an upgraded music generation model capable of creating tracks up to three minutes long—a tenfold increase from Lyria 3's 30-second maximum. The model adds structural music understanding (intros, verses, choruses, bridges) and rolls out to Gemini app paid subscribers, Google Vids, ProducerAI, and enterprise tools including Vertex AI, the Gemini API, and AI Studio.

Reka has released Reka Edge, a 7-billion parameter multimodal model designed for efficient image and video understanding. The model features a 16,384 token context window and is priced at $0.20 per million input and output tokens.

Google DeepMind introduced Lyria 3 Pro, an advanced music generation model capable of creating tracks up to 3 minutes long with structural awareness of musical composition elements like intros, verses, choruses, and bridges. The model is rolling out across multiple Google products including Vertex AI, Google Vids, Gemini app, and the new ProducerAI collaborative tool.

Google has expanded its Lyria 3 music generation model to create tracks up to three minutes long—six times longer than the previous 30-second limit. The new Lyria 3 Pro adds granular control over song structure including intros, choruses, and bridges, and integrates across Google's product ecosystem including Gemini, Vertex AI, and the recently acquired ProducerAI platform.

Opera has rolled out an update to Opera One that allows users to pin Google Gemini to the browser sidebar, eliminating the need to switch tabs when accessing the AI assistant. The update also adds Google Translate to the sidebar and expands split-screen functionality from two to four tabs.

OpenAI has completed pretraining on a new model codenamed 'Spud,' according to an internal memo from CEO Sam Altman reported by The Information. Altman claims the company expects a 'very strong model' within weeks that can 'really accelerate the economy.' To free compute resources, OpenAI will shut down its Sora video generation app.

Anthropic has launched an auto mode for Claude Code that blocks potentially dangerous autonomous actions before execution. The feature, now available as a research preview for Team plan users, acts as a middle ground between constant user oversight and unrestricted agent autonomy.

Anthropic has released Auto Mode for Claude Code, a middle-ground safety feature that automatically executes safe local operations while blocking risky actions like external deployments and mass deletions. A Claude Sonnet 4.6 classifier evaluates each command based on conversation context, and the system reverts to manual approval after three consecutive blocks or twenty total blocks. The feature is available as a research preview for Team plan users, with Enterprise and API access expected shortly.