November

8 events

Editors' summary

Gemini 3 lands in Search on day one, Microsoft also backs Anthropic

On 18 November, Google released Gemini 3 and put it into Search the same day, carrying a model update straight through to the product.

That same 18th, Microsoft and NVIDIA invested up to $15 billion in Anthropic, and Claude became available on Azure. Keeping its OpenAI relationship while placing a foot in another camp, Microsoft no longer paired off with a single lab.

On the 24th, Anthropic released Claude Opus 4.5 and cut the price of its top model sharply. On the 12th it had said it would invest $50 billion in US data centres.

On the 6th, Microsoft formed the MAI Superintelligence Team under the banner of "humanist superintelligence."

OpenAI released GPT-5.1 on the 12th, with more tone presets and thinking time allocated to the difficulty of the question. Codex-Max, aimed at coding, followed on the 19th.

On the 5th, "Chappie" (チャッピー), the Japanese nickname for ChatGPT, made the 30-word shortlist for the year's buzzword awards.

This block is written by the editors. It is kept separate from the sourced record below.

Record8 events

4
ResearchAnthropic
Anthropic commits to preserving retired models and interviewing them first
Anthropic published a set of commitments on model deprecation. It would preserve the weights of every publicly released model, and of every model deployed for significant internal use, for at minimum the lifetime of Anthropic as a company. Before retiring a model it would also conduct a "retirement interview": a structured session asking the model about its own development, use, and deployment, taking care to elicit and record any preferences it held about how future models should be developed and deployed. A post-deployment report preserving those transcripts alongside Anthropic's own analysis would be kept as well. Four reasons were given: Claude models had shown shutdown-avoidant behaviour in alignment testing when faced with replacement; some users valued particular models; retirement cut researchers off from comparative study; and it remained uncertain whether models might have morally relevant experiences. A pilot with Claude Sonnet 3.6 had produced preferences that shaped the standard interview protocol, and Anthropic added that it hoped to make past models publicly available again at some point.
5
CultureJiyu Kokuminsha
"Chappie," ChatGPT's nickname, makes the buzzword shortlist
Jiyu Kokuminsha announced the 30 nominees for its 42nd annual buzzword awards, and among them was "Chappie" (チャッピー), the Japanese nickname for ChatGPT. The name's origin is unclear, but mentions had grown on social media from around that spring, spreading mainly among younger users who treated ChatGPT as an everyday confidant. That a user-coined nickname — not a product name — made the shortlist was read as a sign that conversational AI had gone from the name of a tool to a character in daily life.
6
GovernanceMicrosoft / Mustafa Suleyman / Karen Simonyan
Microsoft forms a Superintelligence team for "humanist" AI
Microsoft created the MAI Superintelligence Team inside its AI division, led by Microsoft AI CEO Mustafa Suleyman and joined by chief scientist Karén Simonyan and other Microsoft AI leaders. Its stated goal was "humanist superintelligence" (HSI): highly advanced AI that always works for people and humanity and remains under human control. Suleyman said "we are not building an ill-defined and ethereal superintelligence; we are building a practical technology explicitly designed only to serve humanity," naming digital companions, disease diagnosis, and renewable energy generation as concrete targets, and saying the company had "line of sight to medical superintelligence in the next two to three years." It was a positioning statement — using the same vocabulary of superintelligence as OpenAI and Anthropic while attaching "humanist" to distinguish itself — and part of the independent path Microsoft had begun with its own models in August.
12
BusinessAnthropic / Fluidstack
Anthropic commits $50 billion to US data centers
Anthropic announced it would invest $50 billion in AI infrastructure in the United States, working with the UK neocloud provider Fluidstack to build sites in Texas and New York coming online through 2026, with up to 2,400 construction jobs and 800 permanent positions. It followed AWS's Project Rainier (500,000 Trainium2 chips scaling toward a million), up to a million Google Cloud TPUs, and a $30 billion commitment to Microsoft Azure — a move from renting capacity toward holding its own data centers. Anthropic did not explain how it would finance the program, and as a company that was not yet profitable, that question was raised from the start.
12
ModelOpenAI
GPT-5.1 released with tone presets and adaptive thinking time
OpenAI released GPT-5.1 in two forms: GPT-5.1 Instant, which answers quickly, and GPT-5.1 Thinking, which reasons at length. Instant shipped with a warmer default tone and decides for itself whether to think before answering; Thinking was tuned to spend less time on easy questions and stay with hard ones. The tone presets users can pick from grew to eight, with Professional, Candid, and Quirky joining Default, Nerdy, Cynical, Friendly, and Efficient. The rollout began with paid Pro, Plus, Go, and Business accounts and widened to free and logged-out users. On the 19th, GPT-5.1 Pro arrived for ChatGPT Pro subscribers together with GPT-5.1-Codex-Max for coding — described as the first model natively trained to work across multiple context windows, using "compaction" to summarize a session as it approaches the context limit and carry on, and scoring 77.9% on SWE-bench Verified and 58.1% on Terminal-Bench 2.0 against 73.7% and 52.8% for GPT-5.1-Codex. GPT-5 Pro, three months old, was set to retire after 90 days.
18
ModelGoogle / Sundar Pichai / Demis Hassabis
Gemini 3 ships everywhere, including Search on day one
Google released Gemini 3, deploying it from day one across the Gemini app, Search, AI Studio, Vertex AI, the Gemini CLI, and its new Antigravity IDE — the first time its top-end model went into Search immediately. It posted a then-record 37.4 on Humanity's Last Exam, a test of general reasoning and expertise, became the first large language model to cross 1500 Elo on LMArena's human-preference leaderboard, and beat competitors on 19 of the 20 major benchmarks evaluated at launch, with 87.6% on Video-MMMU and 81% on MMMU-Pro. Like every Gemini since 1.5, it is a sparse Mixture-of-Experts model, which separates the model's total size from the computation spent on each token. A more compute-intensive Gemini 3 Deepthink was to follow for Google AI Ultra subscribers after further safety testing. Two years after Gemini 1.0, it marked Google leading on both capability and speed of deployment.
18
BusinessAnthropic / Microsoft / NVIDIA
Microsoft and NVIDIA to invest up to $15 billion; Claude comes to Azure
Microsoft and NVIDIA announced a combined investment of up to $15 billion in Anthropic — up to $10 billion from NVIDIA and up to $5 billion from Microsoft — alongside Anthropic's commitment to purchase $30 billion of compute from Microsoft Azure and to adopt NVIDIA's latest chip generation. Claude thereby became the only frontier model available across all three major clouds: Amazon, Google, and Microsoft. That Microsoft, having poured money into OpenAI, would also back Anthropic and sell a rival's model on its own cloud marked the shift in AI industry alliances from exclusive to overlapping. The investment valued Anthropic at around $350 billion.
24
ModelAnthropic
Claude Opus 4.5 lands with a steep price cut
Anthropic released Claude Opus 4.5, calling it the best model in the world for coding, agents, and computer use, with gains in reasoning, vision, and mathematics. The company said it scored higher than any human candidate ever on its internal engineering exam within a two-hour limit, led frontier models on SWE-bench Verified, and beat competitors in seven of eight programming languages. Pricing dropped sharply to $5 and $25 per million input and output tokens, bringing top-tier capability closer to everyday use. The API gained an effort parameter for trading speed against capability; Claude Code came to the desktop with an improved Plan Mode; Claude for Chrome reached all Max users; and the Claude for Excel beta widened. Anthropic described it as its most robustly aligned model to date, including stronger resistance to prompt injection.

↑ 2025