Releases

OpenAI began a limited preview of GPT-5.6 Sol and previewed the Terra and Luna variants.

Mistral AI released Mistral OCR 4, its fourth-generation document-understanding model.

SpaceXAI made Grok models including the cost-efficient Grok 4.3 available through Amazon Bedrock.

SpaceXAI moved Imagine Video 1.5 out of preview into general API availability with improved quality and speed.

Anthropic suspended access to Claude Fable 5 and Mythos 5 after a US government export-control directive concerning model safeguards.

Abridge and NVIDIA announced development of a foundation model purpose-built for clinical conversations using NVIDIA Nemotron and Blackwell infrastructure.

Midjourney made V8.1 the default image model.

Google releases DiffusionGemma as an open diffusion-based model in the Gemma ecosystem.

Cohere established its compact North Small family with North Mini Code for developer workflows.

Cohere launched its first compact model built specifically for developer and coding workflows.

Anthropic released Claude Mythos 5 to a limited set of vetted cybersecurity partners through trusted-access programs.

Anthropic released Claude Fable 5, a Mythos-class model adapted with safeguards for general use across Claude products and the API.

NVIDIA made the 550-billion-parameter Nemotron 3 Ultra open model available for long-running agent workflows.

SpaceXAI released grok-imagine-video-1.5-preview through the API for image-to-video generation.

Microsoft AI announces MAI-Voice-2 Flash as a lower-cost, ultra-efficient speech-generation variant.

Microsoft AI launches MAI-Voice-2 with natural multilingual speech generation and safeguarded short-sample voice adaptation.

Microsoft AI launches MAI-Transcribe-1.5 with high-speed transcription and support for domain terminology across 43 languages.

Microsoft AI launches a lower-cost, latency-optimized Flash variant of MAI-Image-2.5.

Microsoft AI launches MAI-Image-2.5 for high-fidelity text-to-image generation and controllable image editing.

Microsoft AI launches MAI-Code-1-Flash, an efficient agentic coding model integrated into GitHub Copilot and VS Code.

Microsoft AI launches MAI-Thinking-1, its flagship in-house reasoning model for complex instructions, mathematics and coding.

Moonshot previewed an intermediate K2-generation model before Kimi K3.

NVIDIA launched Cosmos 3 as an open frontier world foundation model for physical AI and formed the Cosmos Coalition.

NVIDIA introduced a 34-billion-parameter open reasoning vision-language-action model for level 4 robotaxi development.

ElevenLabs released Dubbing v2 with improved localization and voice preservation.

Anthropic released Claude Opus 4.8 with stronger coding, agentic, computer-use, and professional-work capabilities plus dynamic workflows in Claude Code.

Cognition disclosed the release of SWE-1.6, a fast coding model that became the most-used model in Devin Desktop and could be served at up to 950 tokens per second.

ElevenLabs released its second-generation music model.

Perplexity open-sourced Bumblebee, a security framework for agent systems.

Mistral AI released Mistral Medium 3.5 with improved agentic and enterprise production capabilities.

Cohere released Command A+, a 218B-total/25B-active MoE model for multilingual, multimodal reasoning and agentic work.

Google introduces Gemini Omni, a native multimodal model family designed to accept and generate across multiple media, beginning with video.

Google launches Gemini 3.5 Flash across Google AI Studio, Antigravity and Gemini Enterprise Agent Platform.

Baidu released a more compact flagship model with stronger agents, reasoning and creative abilities.

IBM published the 8B base and instruct Granite 4.1 models.

IBM published the 30B base and instruct Granite 4.1 models.

IBM published the 3B base and instruct Granite 4.1 models.

IBM released Granite 4.1 language, vision, speech, embedding and Guardian models under Apache 2.0.

NVIDIA released an open multimodal model unifying vision, audio and language for efficient agents.

OpenAI discontinued the standalone Sora product as its video strategy shifted to newer experiences and models.

DeepSeek introduced the V4 family through Pro and Flash preview API models.

SpaceXAI released its first named Think Fast speech-to-speech voice-agent model.

OpenAI released GPT-5.5 and GPT-5.5 Pro as upgraded frontier models for complex knowledge work and reasoning.

OpenAI released ChatGPT Images 2.0 with stronger world knowledge, instruction following, detail, and an images-with-thinking mode.

Moonshot released Kimi K2.6 with improved agent capabilities and published model weights.

Physical Intelligence introduced π0.7, a steerable general-purpose robot model with stronger compositional generalization across tasks and embodiments.

Anthropic released Claude Opus 4.7 with improved software engineering, vision, complex multi-step work, and new cybersecurity safeguards.

Baidu released an open-weight 8B DiT text-to-image model.

Google launches Gemini 3.1 Flash Audio for real-time and batch audio understanding and generation.

Midjourney released V8.1 with faster rendering, stronger prompt adherence, and native 2K HD images.