Agents
Claude's Messages API gets message threads, Meituan ships LongCat-2.5, Qodo plugs code review into Codex: the November 3 roundup
Anthropic lets agents send only new messages to the Claude API, Meituan opens LongCat-2.5 for coding agents at $0.30 per million input tokens, and Qodo reviews Codex's work.
Three things for people who build with coding agents: a faster way to run long sessions on the Claude API, a new Chinese model built for agent work, and a review step that now sits inside Codex.
Message threads: send the conversation once
Anthropic has put message threads in public beta on the Messages API, announced on November 2 (Anthropic). Until now, every turn of an agent resent the whole conversation, tool results included. With a thread, the first request carries the full conversation, the system prompt and the tools, plus "thread": {"type": "create"}. Each later request sends only the new messages and the id of the previous response, with "thread": {"type": "continue", "previous_message_id": "..."} and the message-threads-2026-08-12 beta header (docs).
Anthropic says that with threads on by default for most Claude Code users, requests of 500,000 to 1 million tokens see their median time to first token drop by about 35%. Token usage does not change: Claude still reads the whole conversation on every turn, at standard rates and limits.
The limits are clear in the docs. A thread is best-effort: it expires 5 minutes after its last request and is deleted at most 1 hour after its last use. Anthropic says threads are kept in memory, encrypted, and not written to disk. Your app stays the source of truth: keep your own copy of every message, and when a continue request returns a 404 error, send the full conversation again in a new thread. The beta covers every model, on the Claude Platform, Claude Platform on AWS, Microsoft Foundry and Claude models on Google Cloud managed by Anthropic.
Meituan's LongCat-2.5, built for long agent tasks
Meituan released LongCat-2.5, a natively multimodal mixture-of-experts model with 1.6 trillion parameters, about 48 billion of them active per token (LongCat blog). The blog post is dated November 1 and the API changelog lists it as available on November 2, compatible with Claude Code, Codex, Hermes and other agent harnesses (changelog).
The API price is a limited-time discount: $0.30 per million input tokens, $0.006 for cached input and $1.20 per million output tokens, with no end date given (pricing). The context window is 1,048,576 tokens (Vercel AI Gateway).
Meituan's own table gives 88.4 on Terminal Bench v2.1 (70.8 for LongCat-2.0), 38.5 on Terminal Bench v4.0 and 64.6 on DeepSWE v1.1. These scores are self-reported. The weights are not out: GitHub and Hugging Face releases are announced with no date (TechNode).
Qodo reviews Codex's code before the pull request
OpenAI published on November 2 a case study on Qodo, whose plugin is in the Codex marketplace (OpenAI). Before Codex writes code, the plugin can fetch the codebase's rules and the history and patterns of past pull requests. After Codex makes its changes, Qodo reviews them and sends its findings back to Codex in the same session, before the developer opens a pull request. Qodo's CEO Itamar Friedman says some customers have 10,000 or more repositories, where a local change can break a service downstream.
What to keep in mind
- Message threads cut waiting time, not the token bill, and an app must still be able to resend everything.
- LongCat-2.5 is cheap at its launch price, but its scores come from Meituan and the weights are not public.
- The Qodo piece is a case study published by OpenAI, with no independent measurement.
Sources
- Anthropic, "Make long-running agents faster with message threads", checked November 3, 2026
- Claude docs, Message threads, checked November 3, 2026
- Unite.AI, "Anthropic Launches Message Threads Beta for Long-Running Agents", November 2, 2026
- Meituan, "Introducing LongCat-2.5", November 1, 2026, checked November 3, 2026
- LongCat API, change log, checked November 3, 2026
- LongCat API, LongCat-2.5 pricing, checked November 3, 2026
- Vercel AI Gateway, LongCat 2.5, checked November 3, 2026
- TechNode, "Meituan launches LongCat-2.5", November 2, 2026
- OpenAI, "Qodo builds context into AI code review with Codex", November 2, 2026
Spotted an error in this article? Report it · Our corrections policy
