AI & Machine Learning

Alibaba's new flagship tops SWE-bench Pro, Terminal-Bench 2.0 and four other coding leaderboards, but ships as a proprietary hosted model rather than an open-weight release.

Moonshot released Kimi K2.6 under a modified MIT license, claiming parity with GPT-5.4 and Claude Opus 4.6 on coding benchmarks while orchestrating agent swarms that run for half a day unattended.

QUT and Baker-lab researchers built AI-designed allosteric switches that turn on in the presence of small molecules, peptides, or whole proteins and work inside bacteria or on electrodes.

Anthropic's new cloud service handles sandboxing, state, and permissions so companies can ship production agents in weeks instead of months.

OpenAI's April 16 Codex update adds background computer use on Mac, parallel multi-agent execution, persistent memory, and 111 plugin integrations, escalating the agentic coding tool race with Anthropic.

MLCommons' April 2026 inference benchmark round adds text-to-video and vision-language tests while NVIDIA posts a 2.7x jump on DeepSeek-R1 through software alone.

Adobe says its new CX Enterprise Coworker will help large brands automate and personalize customer-experience workflows across its own stack and partner AI platforms.

Opus 4.7 lands across Claude, Bedrock, Vertex AI, and Foundry with unchanged pricing, while Anthropic uses it to trial new cyber safeguards before any broader Mythos rollout.

Equinix has launched Fabric Intelligence, an AI-native operational layer for its 4,400-customer interconnection platform that exposes network provisioning through natural language agents and MCP servers.

Meta releases Muse Spark, a proprietary multimodal reasoning model from its new Superintelligence Labs unit, marking a sharp departure from its open-weight Llama strategy after a nine-month infrastructure rebuild.

Nvidia's new Ising family targets quantum error correction and calibration, claiming 2.5x faster and 3x more accurate decoding than the current open-source standard, with more than 20 labs and hardware makers already on board.

A JAMA Network Open study using the new PrIME-LLM framework finds top AI models excel at final diagnoses with full data but collapse on differential diagnosis when patient information is incomplete.