Google just added native image editing right inside the Gemini app. You can now tweak AI-generated visuals or upload your own photos to swap backgrounds, replace objects, or try new hair colors through simple multi-step prompts. All edited outputs include invisible SynthID watermarks, with a gradual global rollout underway.
Google just expanded NotebookLM's viral Audio Overviews to more than 50 languages, powered by Gemini's native audio capabilities. A new Output Language setting lets you feed it research across Portuguese, Spanish, or English, and have two AI hosts break it all down in your native tongue.
Google just made Meta’s Llama 4 models generally available as a fully managed API on Vertex AI. Developers can now run Llama 4 Scout and Maverick via standard OpenAI-compatible endpoints without provisioning GPUs or managing infrastructure. Why it matters: it brings Meta's flagship open models straight to enterprise cloud pipelines with zero operational overhead.
YouTube is testing an AI-powered search carousel that pulls key clips from videos and plays them directly in search results. The feature does not summarize footage, but extracts the most relevant moments so users do not have to click into full videos. It is currently limited to select Premium users on English queries, raising fresh questions about creator traffic.
Adobe just launched its next-gen Firefly Image Model 4 and 4 Ultra, bringing 2K resolution, sharper text generation, and style matching. It also opened its 1080p video model to everyone and revamped the Firefly web app—which now surprisingly lets users switch to models from OpenAI, Google, and Flux.
Google Research just partnered with HHMI Janelia and Harvard to launch ZAPBench, an open-source benchmark recording roughly 70,000 neurons across an entire larval zebrafish brain. Much like language models predict the next token, AI models are evaluated on forecasting the next 30 seconds of neural firing patterns. Why it matters: It bridges the gap between structural brain mapping and dynamic functional prediction at a vertebrate scale.
Google just upgraded its experimental Music AI Sandbox, rolling out broader access to U.S. creators alongside its new Lyria 2 generation model. The toolkit lets musicians extend tracks, rewrite styles via text prompts, and experiment with real-time generation using Lyria RealTime. All outputs are automatically watermarked with SynthID.
OpenAI just signaled it would be interested in acquiring Google's Chrome browser if a federal judge orders Alphabet to spin it off. ChatGPT product head Nick Turley made the statement during the DOJ's antitrust trial. Meanwhile, Perplexity CEO Aravind Srinivas opposed the divestiture, arguing regulators should focus on opening Android instead.
Korean startup Nari Labs just released Dia, an open-source 1.6B text-to-speech model built by two undergraduates with zero outside funding. The model handles emotional tones, multi-speaker tags, and nonverbal cues like laughter and coughing, outshining commercial leaders like ElevenLabs in benchmark tests. It proves solo builders can now challenge billion-dollar labs.
Google just showed off its Android XR smartglasses in a live public demo at TED. Powered by Gemini, the hardware sports an in-lens color display and can recall where you left a hotel keycard, translate conversations on the fly, and overlay walking directions. Why it matters: It proves multimodal AI is moving to useful everyday eyewear, though consumers likely have to wait until 2026 to buy a pair.
Google just released optimized versions of its Gemma 3 open models using Quantization-Aware Training (QAT). By simulating low-precision operations during training, the 4-bit (int4) weights cut memory demands drastically while slashing perplexity loss by 54%. This means developers can now run powerful 27B models locally on consumer cards like the NVIDIA RTX 3090 instead of expensive cloud clusters.
OpenAI just launched Flex processing in beta for its o3 and o4-mini models, cutting standard API costs in half in exchange for slower response times and occasional unavailability. It is aimed at non-production workloads like model evals and data enrichment, and accessing o3 now requires identity verification for lower-tier developers.
The Wikimedia Foundation just launched an AI-ready dataset on Kaggle to stop scrapers from pounding its servers. Available in English and French, the beta release packages articles into clean JSON with abstracts and infoboxes under Creative Commons licenses, letting developers skip web scraping and messy parsing entirely.
Google just released an early preview of Gemini 2.5 Flash, its first hybrid reasoning model that lets developers turn internal thinking on or off. You can set token budgets from 0 to 24,576 to balance quality, latency, and cost, delivering reasoning performance second only to 2.5 Pro.
Google just opened its Veo 2 video model to developers through Google AI Studio and the paid Gemini API tier. The model turns text or image prompts into eight-second clips with realistic physics and camera controls. It means apps can now generate commercial-grade motion clips without custom video pipelines.
Anthropic just gave Claude an agentic Research feature alongside direct integration with Google Docs, Gmail, and Google Calendar. Claude can now cross-reference your internal company files with live web searches to deliver cited answers in minutes. It is currently rolling out in beta for paid tiers.
Google just unveiled DolphinGemma, a ~400M parameter model developed with Georgia Tech and the Wild Dolphin Project to decode dolphin vocalizations. Operating audio-in and audio-out, it predicts upcoming call patterns much like an LLM predicts words. Best of all, it runs directly on Pixel phones in the wild, with an open-source release planned for summer 2025.
Perplexity's Sonar-Reasoning-Pro just tied Google's Gemini 2.5 Pro for first place on LM Arena's new Search Arena leaderboard. Perplexity models locked down the top ranks by outperforming OpenAI's search models and citing two to three times more sources, winning 53% of direct head-to-head battles against Gemini.
Google DeepMind CEO Demis Hassabis just revealed that Google plans to merge its Gemini foundation models with its Veo video generator to give AI a better grasp of the physical world. Hassabis noted that Veo learned real-world physics by watching heaps of YouTube videos. Combining the two moves Google closer to a true universal digital assistant that actually understands reality.
Google DeepMind just announced it will support the MCP protocol across its Gemini models and SDK. CEO Demis Hassabis called MCP an emerging open standard for the AI agent era, noting Google looks forward to developing it further with the MCP team and the broader industry.
