Meta's Watermelon AI Model Matches OpenAI's GPT-5.5 & Moonshot Kimi K3 Open Weights Arrive & Microsoft Agent Framework Adds Azure Cosmos DB Memory

Meta claims its Watermelon model now matches OpenAI's GPT-5.5, Moonshot's Kimi K3 open weights release arrives Sunday with frontier-class capabilities, and Microsoft enhances its Agent Framework with native memory support powered by Azure Cosmos DB.

AI & ML

Meta's Watermelon AI model now matches OpenAI's GPT-5.5, according to AI superintendent Alexandr Wang in an internal town hall. Wang said the model uses an order of magnitude more compute than Meta's current Muse Spark model and will ship with major gains in coding and agentic capabilities, with plans to close the gap with rival models "pretty soon." Read more
Moonshot's Kimi K3 open weights launch Sunday, July 27 at 00:00 UTC, bringing frontier-class model access for local self-hosting. The model scored first on Arena.ai's Frontend Code Arena in blind developer testing and ranked fourth on the Artificial Analysis Intelligence Index ahead of Claude Opus 4.8, with architectural improvements delivering up to 6.3x faster decoding on long-context tasks. The open-weight release addresses data sovereignty concerns amid geopolitical tensions. Read more

Languages & Runtimes

Microsoft Agent Framework now supports native agent memory via Azure Cosmos DB, enabling durable, cross-session memory for agents built with the framework. The CosmosMemoryContextProvider for Python allows agents to persist and retrieve conversation context across multiple sessions. Read more

Security

OpenAI confirms ChatGPT experienced worldwide outage, though the service has since been restored. Read more

Open Source

OpenSpace enables self-evolving AI agents with skills, Model Context Protocol (MCP), and cost-efficient reuse. The framework requires Python 3.12+ and supports both Anthropic and OpenAI APIs, with a dashboard available via Node.js. Read more
OmniRoute provides free MIT-licensed AI gateway to 290+ providers (90+ free) covering 500+ models including Kimi, Claude, GPT, OpenAI, Gemini, and DeepSeek—compatible with Claude Code, Codex, Cursor, and Cline. Read more

Dev Tools & IDEs

TurboPentest launches affordable self-service AI-powered penetration testing, starting at $99 per target with results in hours instead of weeks. The platform uses AI agents alongside professional security tools to automate pentesting, SAST, SCA, secret scanning, and provide remediation instructions alongside auditor-ready reports. Read more