Mistral AI has launched Mistral Medium 3, an affordable, smaller model that delivers 90% of Claude Sonnet 3.7's performance at 8x lower cost. Medium 3 also outperforms competitors like Llama 4 Maverick in human evaluations, and can be deployed on any cloud environment with at least four GPUs.
The Qwen team has released Qwen3, a new family of open-weight language models comprising two MoE and six dense models. The Qwen3 model family features hybrid thinking capabilities, support for 119 languages, and competitive performance.
Microsoft has introduced three new small language models—Phi-4-reasoning, Phi-4-reasoning-plus, and Phi-4-mini-reasoning—that reportedly deliver complex reasoning capabilities comparable to much larger models while maintaining efficiency for deployment across various computing environments.
AI2's new Olmo 2 1B model outperforms similar-sized offerings from Google, Meta, and Alibaba across multiple key benchmarks, including GSM8K, TruthfulQA, and DROP, while maintaining a small enough size to run on resource-limited hardware.
Amazon Nova Premier, now generally available on Amazon Bedrock, can handle complex workflows requiring multiple tools and data sources. It also features outstanding model distillation capabilities, enabling users to transfer its advanced features to smaller, faster models for production deployment.
FutureHouse has launched a platform featuring four "superintelligent" AI agents—Crow, Falcon, Owl, and Phoenix—designed to help scientists navigate the overwhelming volume of research literature through research capabilities that reportedly outperform both frontier models and PhD-level researchers.
Chinese AI agent startup Manus has secured a $75M investment led by Benchmark, quintupling its valuation to nearly $500M. The funding will reportedly enable Manus to take its AI agent to global markets, including the US, Japan, and the Middle East.
Wikimedia Foundation unveiled a new AI strategy that prioritizes enhancing human contributors rather than replacing them. The strategy focuses on supporting human editors while maintaining Wikipedia's commitment to providing freely accessible knowledge to anyone in the world.
Ghent-based medical imaging startup Nuclivision has secured €5 million in seed funding to commercialize its AI software platform that enables faster, safer, and more affordable PET scans for cancer and Alzheimer's detection.
Meta just launched a standalone Meta AI app powered by Llama 4 that focuses on offering more natural voice conversations and can leverage user data from Facebook and Instagram to provide tailored responses.
Endor Labs has raised $93M to expand its application security platform with new features: the AI Security Code Review, which uses specialized AI agents to detect security issues in pull requests, and an MCP Server that integrates into coding assistants to enable live vulnerability scans.
Datadog has acquired AI-powered data observability startup Metaplane to expand its capabilities in monitoring complex data workflows. Metaplane will be rebranded and will continue to service its customers until integration into the Datadog platform is completed.