Releases

Anthropic released Claude Haiku 5.5, its fastest and most capable small model for high-volume, cost-sensitive workloads. Anthropic says it costs about 75% less per task on average than Haiku 4.5, supports adjustable effort, and is available across the Claude Platform, AWS, Google Cloud, and Microsoft Azure.

AWS released Strands Box, an open-source sandbox for AI agents powered by Dogwood policies. The initial release supports macOS and is designed to package agent runtimes with enforceable controls that can later travel with deployed agents.

AWS introduced the publicly available Physical AI Toolchain, a curated collection of reference architectures, Infrastructure as Code and deployment automation for training and deploying physical AI policies at scale across robotics, autonomous vehicles and smart-factory workloads.

Amazon announced general availability of Anthropic's Claude Haiku 5.5 on Amazon Bedrock and Claude Platform on AWS. AWS says the model is designed for subagents and high-volume, cost-sensitive workloads and costs around 75% less than Claude Haiku 4.5 for most tasks.

SpaceXAI released Grok Bot 0.68.1 with collaborative slide-deck creation and PowerPoint or Google Slides export, formatted email sending, Gmail spam recovery, Slack user-group mentions and faster computer actions.

Microsoft opened preorders for Surface Laptop Ultra and Surface RTX Spark Dev Box, both built around NVIDIA RTX Spark for local AI development and agent workloads. Surface Laptop Ultra availability begins October 16, while the Dev Box is scheduled to ship in the United States in November.

Google expanded SynthID Detector to global English-language availability, letting anyone check whether an image, video or audio file was created with AI from Google or participating partners.

Microsoft made Microsoft Execution Containers generally available on Windows 11, providing policy-driven containment for AI agents with controls over file and network access and integrations spanning Agent 365, Intune and Windows security.

NVIDIA and Microsoft opened preorders for Windows PCs powered by NVIDIA RTX Spark, including systems from ASUS, Dell, HP, Lenovo, MSI and Microsoft Surface. RTX Spark laptops are scheduled to become available on October 16, with compact desktops following in November.

A third-party tracker listed 32 currently open roles at Abridge.

Open IBM roles observed by a third-party hiring tracker.

OpenAI and Atlassian expanded their partnership to deploy GPT-6-family frontier models across Atlassian's platform and Rovo. The agreement also broadens Atlassian's use of Codex and ChatGPT Enterprise and supports deeper integrations with Jira and Atlassian's Teamwork Graph.

Google DeepMind released EmbeddingGemma 2, an Apache 2.0-licensed open-weight multimodal embedding model that maps text, code, images, video and audio into a shared 768-dimensional space. The modular model scales from 270 million parameters for text and code to 740 million parameters for all modalities.

Legora launched Skills so legal teams can encode their standards, processes and preferences for the Legora Agent.

Reflection made the weights for its first model, Beam, publicly available, establishing its open model family.

Reflection released Beam, a 501B-total-parameter sparse mixture-of-experts model with 23B active parameters for coding, reasoning and agentic work.

Amazon made Z.ai's GLM 5.3 available to eligible enterprise customers through fully managed Amazon Bedrock APIs with cross-Region inference, prompt caching and service tiers.

SpaceXAI released Grok Bot 0.68.0 with Marathi, Tamil, Telugu, Swahili and Hausa language options, remote computer removal, verified Gmail send-as support, plugin discovery in the command palette and improved approval and routine controls.

AWS reported that Continuum for code vulnerabilities passed 819 of 920 CyberGym-E2E tasks within the benchmark's 90-minute limit, achieving an 89.0% end-to-end success rate, 23.1 percentage points above the previous public high of 65.9%.

Palantir's public-market capitalization was $455.14 billion based on the latest available closing-market snapshot.

OpenAI says billing for the trusted-access GPT-Rosalind research model begins on October 5, 2026, at $5 per million input tokens, $0.50 per million cached input tokens, and $25 per million output tokens.

Google lists October 5, 2026 as the shutdown date for antigravity-preview-05-2026 in the Gemini API. Developers are directed to migrate to antigravity-preview-09-2026.

Reflection described a planned AI Factory offering that would allow institutions to build localized AI ecosystems around its open-weight models.

Open Meta roles observed by a third-party hiring tracker.

Open roles indexed from Hugging Face's careers site.