LongCat 2.5 Joins AI Gateway, Claude Gets Message Threads + Cloudflare Delays AI Search Billing

AI and SaaS Roundup · 2026-11-03

A new model option for your coding tools, less conversation history to send with each Claude request, and an extra month before Cloudflare starts charging for AI Search. Here’s what’s available and what the small print says.

LongCat 2.5 is now on Vercel AI Gateway

Meituan’s LongCat 2.5 arrived on Vercel AI Gateway on Monday. It accepts text and images, has a 1M-token context window, and is aimed at coding and longer agent workflows. Think working across a codebase with screenshots and documents alongside the code.

You can call it as meituan/longcat-2.5 from the AI SDK, or connect supported coding agents through Vercel’s gateway setup. You don’t need a separate Meituan provider account.

Vercel’s model page currently lists $0.30 per million input tokens and $1.20 per million output tokens. It labels those rates as discounted, without an end date shown, so treat them as today’s price rather than a permanent promise.

Source: Vercel

Claude API message threads mean smaller follow-up requests

If you’re building a chatbot or an agent that keeps calling tools, Claude’s new message threads beta lets the API hold the conversation while you send just the new messages each turn. Anthropic says this reduces request size and time to the first token.

There’s a short memory window: a thread expires five minutes after its last request. Your app still needs its own conversation history so it can replay the chat if the thread has gone. The SDK helper handles that replay while it has the history in memory.

This doesn’t change token pricing or rate limits. It’s also not eligible for zero data retention; Anthropic says the thread is held in encrypted memory and deleted within an hour of its last use. The beta is available on the Claude API and the listed AWS, Anthropic-managed Google Cloud and Microsoft Foundry platforms.

Source: Anthropic release notes

Cloudflare pushes AI Search billing back to December

Cloudflare has moved the start of AI Search billing from 1 November to 1 December 2026. It says you won’t be charged for AI Search usage before then. Useful if you’re still trying out search over your product docs or other content.

This is a delay to billing, not a new permanent free plan. Pricing and the included usage stay the same. It also doesn’t make separate generation or external model-provider calls free.

There’s another change in the same announcement: the default sync interval is moving from six hours to 24 hours. Supported custom intervals still run from 15 minutes to 24 hours, so a docs search that needs fresher updates can use a shorter setting.

Source: Cloudflare