ElevenLabs · Original source
ElevenLabs launched Instant Soundtracks in ElevenCreative, letting users generate a soundtrack and synchronised sound effects for any uploaded video from text prompts. The feature adds music and effects without a traditional scoring pipeline. It extends ElevenLabs' generative audio lineup beyond speech into video soundtrack creation.
Google · Original source
TIME named Google's SynthID one of its Best Inventions of 2026. SynthID embeds imperceptible watermarks in AI-generated images, audio, text and video so platforms and users can detect synthetic media. Google DeepMind developed the system and has rolled it out across Google products and partner tools as generative AI makes digital media harder to authenticate.
Liquid AI · Original source
Liquid AI released d1-ultra, a multimodal decision model it says delivers state-of-the-art performance on hard decision tasks where most models only handle easy, common cases. The company built d1-ultra for constrained action selection across text and multimodal inputs. The launch extends Liquid's decision-model line alongside its on-device LFM family.
Google · Original source
Google reported that magic-state cultivation works on its quantum hardware, addressing a major bottleneck in fault-tolerant quantum computing by producing higher-quality resource states more efficiently. The technique aims to cut overhead needed for useful logical qubits. Google said the result advances the path from noisy processors toward practical error-corrected machines.
Black Forest Labs · Original source
Black Forest Labs released FLUX 3 Image Klein, a 5.6-billion-parameter open-weights model for text-to-image generation, image editing and related tasks. The company said Klein leads its blind evaluations among similarly sized systems while needing less compute than larger FLUX variants. Weights are available so developers can run and fine-tune the model themselves.
NVIDIA · Original source
Adobe is expanding its collaboration with NVIDIA to optimise Photoshop, Premiere, Substance 3D, Frame.io and Brand Concierge on NVIDIA accelerated infrastructure for faster agentic creative workflows. The companies said creators will see higher throughput for generative and editing features that rely on GPUs and CUDA. The partnership targets studio and enterprise production teams.
OpenAI · Original source
OpenAI launched the Rosalind Academic Plan, giving eligible U.S. university labs sponsored access to GPT-Rosalind, its life-sciences model, and the Rosalind workbench for chemistry, genomics and wet-lab workflows. Labs can receive up to 30 Business Standard seats at no cost for up to 12 months after trusted-access approval. Research data is not used for training by default, and API access is not included.
NVIDIA · Original source
NVIDIA released Nemotron 3.5 Super VL, an open 120-billion-parameter vision-language model it says delivers the fastest task completion and highest accuracy for specialised always-on agents, plus a Nano VL variant for efficient desktop and edge use. The models extend the Nemotron family from data centre to PC. NVIDIA positioned the release for enterprises building agentic systems that need multimodal reasoning at lower cost.
Salesforce · Original source
Salesforce said Tottenham Hotspur has launched Ask Spurs, a multilingual AI agent built on Agentforce that answers fan questions about ticketing, memberships, retail and stadium events in eleven languages on the club's website and app. The club has unified more than 4.6 million fan accounts into SpursID profiles using Salesforce Data 360 integrated with Snowflake. Ask Spurs has an 86% engagement rate, with 20% of interactions resolved by Agentforce alone and unresolved questions handed to human support staff.
Scale AI · Original source
Scale launched Scale Post-Training, a new service within its GenAI Portfolio that helps enterprises customize open-weight models, including NVIDIA's newly launched Nemotron 3.5 Super VL and the wider Nemotron family. Scale said it had early access to the model and post-trained it ahead of launch with enterprises including Skydance and EY. In a Skydance finance workflow, the post-trained model matched the accuracy of the frontier model used in production at about 40x lower inference cost and 5x faster response times.
NVIDIA · Original source
NVIDIA joined the European Commission's AI in Science Enablers Initiative, launched at the AI in Science Summit in Dublin and connected to RAISE, the Resource for AI Science in Europe. NVIDIA's pledge covers access to accelerated tools, models and platforms, compute and cloud resources, skills and training, research collaborations and startup support. The company noted it has released more than 250 open models, 250 open datasets and over 1,400 open source software projects, and its Deep Learning Institute launched an open courseware hub.
Amazon · Original source
Amazon introduced two new devices for Bee, its personal AI that takes note of conversations and turns them into summaries, to-dos, reminders and insights. Bee Frontier is a wearable for the wrist or clothing with up to seven days of battery life, available now for $74.99. Bee Eclipse attaches magnetically to a phone and will be available soon from $99.99. Bee can now connect to Alexa+ and Amazon.com, and a Bee Pro subscription costs $11.99 a month.
Cohere · Original source
Cohere announced a proposed agreement to acquire AISLE, a company founded in 2025 that builds autonomous systems to discover and help fix previously unknown software vulnerabilities. AISLE says it found more than 500 CVE-assigned vulnerabilities across over 100 codebases in its first year, including all 12 in OpenSSL's January security release. Cohere said the deal would strengthen security across its sovereign AI platform and deepen its presence in Europe, following its acquisition of Reliant AI and partnership with Aleph Alpha.