Model Releases/Updates
AWS Machine Learning BlogFirst-partyClaude Haiku 5.5 Launches on Amazon Bedrock and Claude Platform on AWS
AWS announced that Claude Haiku 5.5 is available on Amazon Bedrock and Claude Platform on AWS.
Anthropic Releases Claude Haiku 5.5
Anthropic released its latest model, Claude Haiku 5.5. According to the @claudeai content it cites, this is its cheapest, fastest, and most capable small model to date, with an average running cost about 75% lower than Claude Haiku 4.5.
Liquid AI Open-Sources d1-3B and d1-omni-600M On-Device Decision Models
Liquid AI released two open-weight models in the d1 decision model family, d1-3B and d1-omni-600M (experimental).
Product Releases/Updates
NVIDIA and Microsoft Release RTX Spark and Windows Agent Infrastructure
NVIDIA and Microsoft announced at a Windows AI event in San Francisco that they co-designed software and hardware for the agentic era of Windows PCs, and Microsoft announced the general availability of OS-level infrastructure Microsoft Execution Containers (MXC), allowing agents to run securely and persistently in the background under system control.
Microsoft Research open-sources Agent Lightning v1.0: a lightweight 3,500-line agent RL framework supporting training with real harnesses
Microsoft Research has open-sourced Agent Lightning v1.0, a lightweight agent RL framework of about 3,500 lines of code that proposes the Harnessed Agentic RL paradigm, allowing the same agent harness used at deployment time to participate directly in reinforcement learning without reimplementing the agent inside the training framework.
Industry News
GitHub Blog · AI & MLFirst-partyGitHub launches a ModernBERT-based secret detection model, expanding push protection
GitHub released a ModernBERT classifier fine-tuned jointly with Microsoft Applied Sciences that uses code context to identify unstructured secrets, can evaluate a batch of candidate secrets within 2 milliseconds, and is said to more than double the number of secrets that can be blocked.
Paper Research
After fine-tuning, Nemotron reaches gold medal level in both IOI 2026 and IMO 2026
Starting from Nemotron 3, the NVIDIA team used SFT, RL, and a generate-verify-improve inference pipeline to build a specialized model that reaches the gold medal threshold in both IOI 2026 and IMO 2026.
Google ResearchFirst-partyGoogle Research's three-month patent drafting experiment: AI assistance boosts output, but does not accelerate junior lawyers' growth in judgment
Google Research published a three-month field experiment on patent drafting at NBER, in which the then-unreleased Google Labs AI patent writing assistant (now part of Gemini Notebook) was randomly made available to 133 lawyers at 11 intellectual property law firms, with two-thirds receiving early access.