1. OpenAI releases GPT-5.5 and GPT-5.5 Pro
OpenAI released its GPT-5.5 and GPT-5.5 Pro models, featuring a 1 million token context window and optimized agentic capabilities. The new models match the latency of GPT-5.4 while reducing token usage by approximately 40% for coding tasks. API pricing is set at $5 per 1 million input tokens and $30 per 1 million output tokens. The models are available immediately in ChatGPT and Codex, with API access rolling out shortly.
2. OpenAI rotates macOS app signing certificates following supply chain attack
OpenAI is revoking and rotating the signing certificates for its macOS applications, including ChatGPT Desktop and Codex, following a supply chain compromise of the Axios developer library. A malicious version of Axios was executed during OpenAI's GitHub Actions workflow, potentially exposing notarization material. Developers and users must update their macOS applications by May 8, 2026, after which older versions will no longer function.
3. Researchers identify undisclosed native messaging bridge in Claude Desktop
Security researchers discovered that the Anthropic Claude Desktop application for macOS silently installs a Native Messaging manifest file across Chromium-based browsers. This bridge pre-authorizes specific browser extensions to communicate with a local binary running outside the browser's sandbox with user-level privileges. The installation occurs without explicit user consent and persists even for browsers not currently installed on the system. This behavior bypasses standard browser trust models and introduces potential unauthorized system access risks.
4. Anthropic resolves Claude Code latency issues and resets usage limits
Anthropic deployed a fix for Claude Code, the Claude Agent SDK, and Claude Cowork to resolve latency issues that caused the user interface to appear frozen. The degradation was traced to a recent change that defaulted the reasoning effort from high to medium for Sonnet 4.6 and Opus 4.6. Anthropic reverted this change, allowing developers to default to higher intelligence and manually opt into lower effort for simpler tasks. Usage limits for all subscribers have been reset as of April 23.
5. Xiaomi releases MiMo-V2.5 and MiMo-V2.5-Pro multimodal models
Xiaomi released the MiMo-V2.5 and MiMo-V2.5-Pro open-source models, designed for complex software engineering and long-horizon autonomous tasks. The models feature a 1 million token context window and native multimodal support for text, image, audio, and video inputs. The Pro variant is optimized to sustain over 1,000 sequential tool calls. These releases provide developers with high-capability agentic models at a lower token cost compared to proprietary alternatives.
6. Alibaba releases Qwen3.6-27B dense language model
Alibaba released Qwen3.6-27B, a 27-billion-parameter dense language model optimized for agentic coding and repository-level reasoning. The model features a native 262,144-token context window that can be extended to over 1 million tokens. It introduces a "Thinking Preservation" feature to maintain reasoning context across multi-turn interactions. The model is available under the Apache 2.0 license on Hugging Face and ModelScope.
7. Meta releases Sapiens2 vision transformers
Meta released the Sapiens2 family of high-resolution vision transformers on Hugging Face. Pretrained on 1 billion human images, the models are optimized for human-centric perception tasks such as pose estimation, body-part segmentation, and surface normal estimation. The release includes models ranging from 0.4 billion to 5 billion parameters, supporting native 1K resolution and hierarchical variants for 4K inputs.