Releases

Cohere introduced Parse for enterprise document intelligence across complex multimodal files.

Google releases Gemini Omni 1.1 Flash for native real-time multimodal generation and editing.

Google launches Gemini 3.5 Transcribe for accurate, multilingual speech-to-text workloads.

DeepSeek launched an experimental native multimodal V4 Flash model.

Harvey introduced Tenet, its first post-trained open-weight model for long-horizon legal reasoning.

Skild released S1, an in-context robotics foundation model that can execute unseen tasks from a single video demonstration without fine-tuning.

DeepSeek rolled out the GA V4 Pro model across app, web and API with stronger production agent capabilities.

Google launches Gemini 3.7 Flash, improving intelligence and multimodal performance while retaining low latency.

SpaceXAI released Grok 4.6 with a focus on long-running agents and ambitious interactive and visual work.

Microsoft AI announces MAI-Image-2.6 with major benchmark improvements in generation and image editing.

OpenAI introduced GPT-5.6 Cyber for high-end defensive cybersecurity use through Daybreak.

SpaceXAI released Imagine Image 2.0 for precise image generation and editing with stronger text rendering.

Anthropic updated Claude Fable 5's biology safeguards as part of its safety work for high-capability models.

DeepSeek released the official Flash API in public beta with upgraded coding and agent performance.

SpaceXAI added text, image and voice references plus native 1080p support to Imagine Video 1.5.

Thinking Machines released Inkling-Small, a 276B-total-parameter mixture-of-experts model with 12B active parameters and performance comparable to the larger Inkling model.

Runway retired Gen-3 Alpha Turbo and first-generation Aleph API identifiers.

Runway released Aleph 2.0 for advanced video transformation and professional output workflows.

SpaceXAI released its second-generation real-time speech-to-speech model.

Moonshot published the full Kimi K3 weights and implementation resources.

Midjourney retired the V8.0 alpha after V8.2 became available.

Midjourney launched its V8-era dedicated Edit Model.

Midjourney released V8.2 as the default model with improved aesthetics, image quality, personalization, and a new Edit Model.

Anthropic released Claude Opus 5 with improvements in coding, long-running agents, professional work, and scientific reasoning.

Cursor launched Router, a model-routing system that dynamically selects an appropriate coding model.

SpaceXAI expanded Grok 4.5 availability across grok.com, X, iOS and Android.

Google launches Gemini 3.6 Flash with stronger reasoning and agentic performance at Flash-class latency.

Google DeepMind introduces Gemini 3.5 Flash Cyber for defensive cybersecurity analysis and agentic workflows.

Google launches Gemini 3.5 Flash-Lite for cost-efficient production workloads.

Kimi K3 launched with native vision, agentic capabilities and a one-million-token context window.

Moonshot introduced Kimi K3, its 2.8T-parameter native multimodal frontier model family.

SpaceXAI introduced Grok 4.5 as its strongest model for coding, agentic tasks and knowledge work, trained alongside Cursor.

Synthesia announced that its latest EXPRESS-3 model powered the new Interactive Avatar API's real-time photorealistic and style avatars.

Thinking Machines published Inkling's full weights and made the model available for customization through Tinker.

Thinking Machines released Inkling, a 975B-total-parameter multimodal mixture-of-experts model with 41B active parameters and up to a one-million-token context window.

SpaceXAI open-sourced the Grok Build coding-agent harness and terminal user interface.

NVIDIA introduced an edge-optimized Cosmos 3 model for real-time physical AI systems.

Cohere Labs documented the Tiny Aya family and community applications for compact open multilingual AI.

Meta released Muse Spark 1.1 and made it available through Meta Model API.

OpenAI launched GPT-5.6 Sol, Terra, and Luna broadly across ChatGPT, Codex, and the API.

Cognition released SWE-1.7, its then-most-capable coding model, optimized for long-horizon agentic tasks and served at up to 1,000 tokens per second.

Mistral AI released Robostral Navigate, a vision-language-action model for robotic navigation.

OpenAI released GPT-Live-1 and GPT-Live-1 mini for low-latency, natively multimodal voice and realtime interaction.

Cohere released an open speech-recognition model tailored to Arabic languages and dialects.

Meta announced Muse Video, a generative video model with native audio support, as coming soon.

Meta released Muse Image for instruction-following image generation, precise editing and multi-reference composition.

Mistral AI released Leanstral 1.5 with improved formal reasoning and theorem-proving capabilities.

Anthropic restored global access to Fable 5 and resumed restricted access to Mythos 5 after the relevant export controls were lifted.

Anthropic released Claude Sonnet 5 as an upgrade to Sonnet 4.6 for agentic work, coding, and professional tasks.

Meta released training code for Brain2Qwerty v1 and v2; v2 performs real-time non-invasive sentence decoding from brain recordings.