Signal / Noise (Nov 2, 2026): ChatGPT and Codex Share Limits and AWS Grows 39%

Four items you can act on this week, one promise to watch and one claim to handle with care. If you run sales, logistics or retail, start with the first two: they are about the cost and speed of the AI tools your people already use. This digest covers AI news for November 2, 2026, and we have not tested any product named here.

How we label items: Signal means an actionable development, Watch means a credible change that is not yet broadly usable or sufficiently validated, and Noise until reviewed means a claim that has not been independently reviewed. Each number belongs to whoever published it, and we read each item at its primary source on November 2.

Signal: ChatGPT and Codex will share one usage allowance from December

OpenAI's Help Center says that starting in December, the usage allowance of the Plus, Pro and Business plans will be shared across ChatGPT and Codex, as it brings the agentic capabilities of its Work feature into the main ChatGPT experience. Usage will be measured in tokens, how much of the allowance a task uses depends on the model and the task, and Instant chat will not draw from it. Remaining usage and reset times will show in Settings under Usage, and at the limit a user can wait for the reset, upgrade or buy credits. OpenAI says it expects most users will not wait for a reset more often than they do now. The article does not give the size of the combined allowance or a specific start date in December.

Builder take: if a team's workflow depends on a subscription allowance, ask what happens to a long job when the allowance runs out halfway, and keep a margin or an API route in the budget. For operators: ask whoever builds your software whether their coding tools run on seat plans, and what the plan is when limits change.

Signal: AWS grew 39% in the third quarter

Amazon's third-quarter release, which covers the quarter ended September 30, 2026, reports AWS net sales of $45,984 million, up from $33,006 million a year earlier. Amazon calls that its fastest AWS growth in nearly five years. AWS operating income was $17,773 million, about 38.7% of sales by our arithmetic, and Amazon cites a run rate of $184 billion, an annualization of recent sales rather than revenue received over a year. On AI, the release says Fortune 500 companies are spending nearly seven times more on Amazon Bedrock than a year ago, that total Bedrock token volume grew nearly 17 times, and that AgentCore has tens of thousands of customers. It also says Amazon expects 2026 cash capital expenditures of about $220 billion for the company as a whole. The Bedrock multipliers describe usage and spending in specific categories, not the whole AI market, and they compare different measures. These are Amazon's own figures.

Builder take: usage of managed model and agent platforms is growing faster than the cloud as a whole, in Amazon's telling, so ask where your agent runs and what spend and permission controls that platform provides. For operators: this is a vendor's report, not a verdict on any product; use it as a prompt to ask your supplier which cloud and which controls they use.

Signal: Microsoft says North Korean actors use AI, with people in charge

On October 29, Microsoft Threat Intelligence published an assessment of AI use by North Korean threat actors. It rates their overall adoption of AI as high and says AI use supports their existing goals. In its words, AI-enabled operations remain human-directed and have not become fully autonomous. The post describes AI help with reconnaissance and tailored lures, with fake résumés and personas, with writing and debugging malicious code and with scaling repetitive tasks. It reports that more than 100 PDF lure files attributed to a group Microsoft calls Onyx Sleet were created in August and September, with metadata indicating rapid batch creation that Microsoft assessed as highly likely to involve AI, that two groups have used an open-source agentic penetration-testing framework since late October, and it advises employers to verify the identity and digital footprint of job candidates. The post names no model vendors, though it mentions an AI coding tool.

Builder take: the reported pattern is faster preparation of familiar attacks, with people still directing them. For operators: hiring is one of the paths Microsoft highlights, so ask your recruiting and contractor processes how identities are checked.

Signal: Apple Intelligence is coming to China, with Alibaba's Qwen in parts

Apple's China newsroom release dated October 31 says Apple Intelligence features are scheduled to roll out to supported iPhone users in China, with testing starting on November 4 in the latest iOS developer beta and a public rollout planned for December with iOS 27.3. Alibaba's Qwen supports parts of it, including some Siri requests, questions about photos and documents, and creating original content in Writing Tools. The release says Siri AI and the other next-generation Apple Intelligence features announced at WWDC 2026 are not part of this phase, that other platforms will follow after necessary approvals, and that core features support English and Simplified Chinese as the language of the device and Siri.

Builder take: this is a consumer-device rollout with a named model partner, so it matters mainly if your product reaches iPhone users in China. For operators: nothing to do unless you serve that market.

Watch: Moonshot promises Kimi K3.1 Pro weights by November 4

Moonshot's quick-start documentation says Kimi K3.1 Pro is the first model in its K3.1 series, has a 1M-token context window, and that its full weights will be released by November 4. It describes the model as its most capable 3T-class open-weight model to date, and says the model is already available through its API. The page names no license for K3.1 Pro, and as of November 2 the Hugging Face page for Moonshot's organization lists no K3.1 repository; the newest entry is the earlier Kimi K3 model. The earlier model is under the Kimi K3 License, and we do not know whether the new one uses it.

Builder take: wait for the files and the license text before planning around it. For operators: a promise is not a release; ask any vendor which version of a model they actually run.

Noise until reviewed: Kimi's performance claims are vendor-reported

Moonshot's launch post reports 65.7% for Kimi K3.1 Pro on Terminal-Bench 4.0 and 61.2% on its own internal Kimi Code Bench 3.0 Long-horizon. Its footnotes say the K3.1 Pro runs used the Claude Code harness and that competitors' Terminal-Bench numbers are the maximum from their own posts or the official leaderboard, with one taken from its release blog, so the table is not a uniform head-to-head test. In Moonshot's own table the model sits narrowly behind one competitor on Terminal-Bench 4.0. These are published figures, not independent validation. Builder take: use them as a screening signal and test the model on your own tasks, comparing cost per completed task, reliability and speed. For operators: ask suppliers which model version they run and whether their performance claims come from independent testing.

What to do this week

One habit for all five: ask who measured each number. Today's longer reads are our decode of CVE-2026-103572 and our teardown of the dsh-superagent research prototype. For earlier context see the October 30 edition, our comparison of AI across clouds and our coding agents comparison.

Sources

• OpenAI Help Center: Understanding ChatGPT and Codex usage limits, read November 2, 2026
• Amazon, October 29, 2026: Amazon.com Announces Third Quarter Results
• Microsoft Threat Intelligence, October 29, 2026: Tracking AI adoption in North Korean threat actor tradecraft
• Apple Newsroom (China), October 31, 2026: Apple Intelligence features are coming to users in China
• Moonshot AI: Kimi K3.1 Pro quick-start documentation, read November 2, 2026
• Moonshot AI: Kimi K3.1 Pro launch post, read November 2, 2026
• Hugging Face: moonshotai organization page, read November 2, 2026