Anthropic's beta keeps thread context across a full Slack conversation when you @-mention it. For teams already on Slack plus Claude, it is the first native always-on context integration, aimed straight at the Glean and Guru category.

The cyber-tuned model scores 85.6 percent on CyberGym versus 81.8 percent for base GPT-5.5, per OpenAI's Daybreak announcement. OpenAI launched Patch the Planet with Trail of Bits across 30-plus open-source projects. This is the first capability-gated, defenders-only release OpenAI has shipped at this scale. Read our take.

Identity verification by July 8 is now the prerequisite to restore Fable 5 access, the first hard date in the Fable saga. If you build on Fable 5, you need a verification plan this week.

Full-page web extraction with a new Research Index spanning 3M-plus arXiv papers, built for agent and RAG pipelines moving off Tavily or Jina. We broke down the tradeoffs in our web-search API comparison. Try Firecrawl →

Cursor acquired Continue on June 16, and users have until July 15 to export their data. Decision matrix: BYOK and local users go to Cline, JetBrains shops to Copilot or Cline, and anyone who wants the most capable AI editor goes to Cursor. Do not wait on the export.

Firecrawl, Tavily, Exa, and Jina compared on what each does best, so you pick by workload instead of marketing. Read it. Try Firecrawl →

The tab-prediction model is still the reason to pay, even after the SpaceX acquisition reshuffled the roadmap. Our verdict on who should switch and who can wait.

How an exposed Sentry MCP key lets an attacker run code through Claude Code or Cursor, and the config change that closes the hole.

Kimi K2.7 leads MCP tool-use accuracy, DeepSeek V4 Pro leads SWE-bench. Which open model fits your workload depends on which of those you weight.