Frontier
Model launches, benchmarks, multimodal systems, and eval wars.
Monitor
OpenAI is discontinuing its Sora video generation platform amid compute reallocation, leading Disney to scrap a planned $1B investment that would have allowed AI-generated content from its IP. The move highlights challenges in scaling generative video tech.
Cursor publishes in-depth report on training its frontier coding model Composer 2, including CursorBench benchmark, custom kernels, and RL scaling. Emphasizes agentic coding efficiency.
OpenAI is discontinuing its Sora app and API after initial hype faded amid deepfake concerns and low adoption; the move impacts a planned $1B Disney deal and signals a pivot away from standalone video tools.
Over 30 major AI models launched in Q1 2026 alone, including GPT-5.4, Claude 4.6 variants, Gemini 3.1, and Qwen3.5, signaling unprecedented iteration speed.
OpenAI CEO Sam Altman teases a powerful new model that could 'accelerate the economy,' with deployment team rebranded to 'AGI Deployment.' Expectations are high following recent GPT-5.4 successes.
xAI's Grok Imagine debuted at #1 overall on Design Arena's video benchmark, achieving Elo 1342 with 58.9s average generation time, setting new standards in preference and speed.
Anthropic rolled out auto mode in Claude Code, allowing the AI to handle file writes and commands independently while classifiers block risky actions. Available as research preview on Team plan, with Enterprise rollout soon.
New engineering blog post reveals how Anthropic uses multi-agent systems to extend Claude's capabilities in complex software tasks. Focuses on autonomous frontend dev and sustained app building. Builds on recent computer-use advances.
Claude Code and Claude Cowork now control user computers to complete tasks autonomously, including browser navigation, app use, and file saving, with new 'auto mode' for smart permission decisions and safeguards. Anthropic positions it as a research preview for agentic workflows. Developers rave about its potential to automate daily grinds.
Mistral's new Forge platform lets enterprises build custom frontier-grade AI models grounded in proprietary data, targeting business-specific intelligence.
Xiaomi's trillion-parameter MiMo-V2-Pro model has skyrocketed to the top of OpenRouter rankings, processing 500B tokens daily and outperforming Google models in agentic tasks. The stealth release underscores China's rapid AI ascent.
Anthropic rolled out 'computer use' for Claude, allowing the AI to interact with desktops, browse, type, and execute tasks autonomously to rival viral agents. This agentic upgrade positions Claude as a stronger contender in practical AI applications.
A new benchmark-topping model called Hunter Alpha emerges, speculated to be DeepSeek's upcoming V4 from China, stirring talk of closed-source powerhouses challenging open leaders.