1. Amazon Bedrock adds OpenAI models in limited preview
AWS and OpenAI have partnered to bring OpenAI's frontier models, including GPT-5.5, to Amazon Bedrock in a limited preview. The launch includes OpenAI models, Codex, and Amazon Bedrock Managed Agents powered by OpenAI. This integration allows enterprise developers to access OpenAI capabilities directly within their existing AWS environments. The deployment utilizes standard AWS security, governance, and operational controls.
2. Wiz Research discloses GitHub RCE vulnerability
Wiz Research has uncovered CVE-2026-3854, a remote code execution vulnerability in GitHub's internal git infrastructure. The injection flaw allows authenticated users to execute arbitrary commands on backend servers using a standard git push command. GitHub mitigated the issue on GitHub.com within six hours and released patches for all supported versions of GitHub Enterprise Server. Administrators of GitHub Enterprise Server must upgrade immediately to prevent full server compromise and access to hosted repositories.
3. GitHub shifts Copilot to usage-based billing
GitHub will transition its Copilot AI service to a usage-based billing model starting June 1. Subscribers will receive a monthly allotment of AI credits matching their subscription payment, with additional usage billed based on token consumption. Pricing for excess usage will be calculated using listed API rates for input, output, and cached tokens. Simple suggestions like code completion remain free, but Copilot code reviews will consume GitHub Actions minutes.
4. DeepSeek reduces V4-Pro API pricing
DeepSeek has implemented a 75 percent promotional discount on its V4-Pro model pricing until May 5, 2026. The reduction brings the input price down to approximately $0.036 per million tokens. The company also permanently reduced the cost of input cache hits by 90 percent across its entire API suite. These changes lower the operational costs for developers deploying DeepSeek models in production environments.
5. Xiaomi open-sources MiMo-V2.5-Pro model
Xiaomi has released the MiMo-V2.5-Pro, a 1.02 trillion-parameter Mixture-of-Experts model, under the MIT license. The model features 42 billion active parameters and supports a one-million-token context window. It is specifically optimized for complex agentic workflows and software engineering tasks. The permissive license allows developers to use the model for commercial applications and secondary training.
6. Nvidia releases Nemotron 3 Nano Omni model
Nvidia has launched Nemotron 3 Nano Omni, a 30-billion parameter multimodal model that unifies text, vision, and speech processing. The model utilizes a hybrid Mixture-of-Experts architecture combined with vision and audio encoders to eliminate separate perception modules. It is designed for low-latency agentic applications, allowing agents to rapidly interpret full HD screen recordings and voice activity. The model is available as an Nvidia NIM microservice and can be deployed on local consumer hardware.
7. Poolside releases Laguna agentic coding models
Poolside has released Laguna M.1 and Laguna XS.2, two new models optimized for long-horizon software engineering tasks. Laguna M.1 is a proprietary 225-billion parameter Mixture-of-Experts model designed for enterprise environments. Laguna XS.2 is a 33-billion parameter model released under an Apache 2.0 license, allowing developers to run it locally on a single GPU. The company also released "pool," a research preview of its internal agent harness used for reinforcement learning training and evaluation.
8. Mistral AI launches Workflows orchestration engine
Mistral AI has released Workflows in public preview as part of its Studio platform. The product is a production-grade orchestration layer powered by Temporal. It is designed to help organizations move enterprise AI systems from isolated proofs of concept into reliable, scalable business processes. The engine manages the infrastructure required to run complex AI models and agentic tasks reliably at scale.
9. OpenAI releases Symphony agent orchestration specification
OpenAI has published Symphony, an open-source reference specification for orchestrating Codex coding agents. The system turns project-management boards like Linear into control planes, allowing agents to continuously pick up tasks from an issue tracker. It manages separate workspaces, monitors continuous integration, and prepares pull requests for human review. Symphony is provided as a reference implementation for engineering teams to study, fork, or rebuild within their own infrastructure.
10. Red Hat releases Tank OS for OpenClaw deployments
Red Hat has released Tank OS, an open-source tool for securing enterprise OpenClaw agent deployments. The software runs OpenClaw agents inside isolated, rootless Podman containers on Fedora Linux. This architecture prevents agents from interfering with the host machine or other agents while securing credentials. It allows IT teams to manage fleets of AI agents using standard container management practices.
11. Cua releases background computer-use driver for macOS
Cua has released Cua Driver, a background computer-use driver for macOS 14 and later. The tool allows AI agents to click, type, scroll, and read native applications without stealing the user's cursor or keyboard focus. It provides a command-line interface that can be scripted or called from coding agent shells. This enables developers to run GUI-operating agents concurrently on a host machine without interrupting human workflows.
12. Claude CLI bug causes subagent code editing refusals
A regression in the Claude CLI version 2.1.111 is causing subagents to refuse legitimate code edits. The bug injects a malware analysis reminder into every Read and Grep tool result, overriding user instructions. This prompts models like Opus 4.7 to halt exploratory file reads and refuse to augment code in standard repositories. The issue persists despite being marked as fixed in an earlier version, blocking parallel agent workflows for developers.
13. Warp open-sources terminal client
Warp has open-sourced its terminal client repository, licensing the core client under AGPL v3 and the UI framework under MIT. The company is utilizing an agent-first workflow managed by its cloud orchestration platform, Oz, to handle community contributions. The update also introduces support for a wider range of open-source models, including Kimi, MiniMax, and Qwen. Developers can now build and run the Warp client from source or customize their terminal experience programmatically via a new settings file.