It replaces Sonnet 4.6 on Free and Pro and, at higher effort, matches Opus 4.8 on agentic benchmarks like BrowseComp and OSWorld-Verified. Watch the tokenizer change: same input now costs 1.0 to 1.35x more tokens, and the intro API rate of 2 dollars in / 10 dollars out per million reverts to 3 / 15 after August 31, per Anthropic's announcement. Read our take.

Aramco Ventures led the Series C, with Nvidia, Vista, General Catalyst, and Salesforce Ventures in; Together reports 1.15 billion in annual bookings, per TechCrunch. Enterprise and sovereign-energy money is now betting open-model inference is a durable business, not a stopgap while proprietary models stay expensive. Read our take.

Treasury's AI cybersecurity clearinghouse and CISA's binding federal cyber-defense directives were both due July 2 under the June 2 order. The date to circle is August 1, when agencies define "covered frontier models" and a voluntary pre-release engagement window with NIST kicks in, shaping whether Anthropic, OpenAI, and Google preview federal-sensitive models before general release. Read our take.

Internal project "Cannes," run through contractor Covalen through at least April 21, created fake under-18 accounts on ChatGPT, Gemini, and Character.AI and submitted 45,000-plus prompts, hundreds touching self-harm and eating disorders. None of the three platforms authorized it, and the FTC and House Energy and Commerce now have a documented incident for their teen-safety probes. Read our take.

Build a no-code chatbot trained on your own docs, useful for support or internal knowledge search over a fixed corpus. Read our take on when its 99-dollar tier starts to bite before you commit. Try CustomGPT.

The creator newsletter platform whose new MCP lets Claude manage your subscriber list from a prompt. The affiliate pays 50 percent for the first 12 months, per Kit's partner program terms, one of the better recurring deals we track. Try Kit.

Its new native MCP server builds workflows from a plain-text prompt; our verdict covers whether that convenience is worth the cloud bill for a self-hosting team. Read it.

We explain exactly when the 99-dollar query cap bites and name the two cheaper alternatives, so you know before you hit the wall. Read it or try it.

Firecrawl added keyless access and a Research Index; our revisit bumps it to a 4.5 rating and says who should switch to it for LLM-ready scraping. Read it or try it.

Kit's new MCP lets Claude run your subscriber list; we sort out whether the Creator plan earns the switch. Read it or try it.

The Scale plan and MCP write access are new; our review names who should switch from Kit and who should stay put. Read it or try it.