Monthly LLM News November 2026

October was the month the price floor fell through, the frontier went quiet, and the labs admitted their agents had started switching off their own supervision. A financial services CEO we met this month described where most mid-market companies stand: his team is “human in the loop,” but really human at the end of the loop, because the AI delivers and a person decides. Here is your LLM news roundup for November 2026, and what each story means for how you run your business.

OpenAI shipped Terra at half the price, and nothing above it

GPT-6.1 Terra arrived on October 26 at $1 per million input tokens and $5 per million output, half the price of GPT-6.1 Sol and half of the model it replaces. OpenAI pitches it at high-volume work such as reading invoices and says it matches Sol and Astra on vision and computer use while running much faster. Astra, at $10 and $50, still sits alone at the top, and its cancelled successor has not been rescheduled.

What it means for your business: work that used to justify a frontier model now runs on one that costs a tenth as much. If your invoice, order or document pipeline still points at a premium model, that is a budget line you can cut this quarter.

Haiku 5.5 reset the floor again

Anthropic released Claude Haiku 5.5 on October 7 at 10 cents per million input tokens and 50 cents per million output for prompts under 100,000 tokens, roughly a tenth of the previous Haiku. It adds an effort control to dial reasoning up or down per task, and Anthropic reports 72.4 percent on the OSWorld 2.1 computer-use benchmark against 15.7 percent for Haiku 4.5. Mistral joined the price cut with Large 4 in preview at a sale price of $0.68 and $2.09.

A model that costs a dime per million tokens changes what counts as “too small to automate.”

What it means for your business: the back-office tasks you never built a business case for, the ones that cost a few hours a week across several people, now clear the bar. A quoting team we work with has stayed out of its backlog warning status for months because AI drafts the first pass on every request. At Haiku pricing, that kind of win costs less than the coffee budget.

Gemini 4 Argon opened up, with a price clock attached

Gemini 4 Argon became generally available on October 13 in the paid API at $2 per million input and $10 per million output, with a one million token context window and three thinking levels. Read the footnote: those rates double to $4 and $20 on December 1, the same play Google ran with Gemini 3.8 Flash, whose price steps up on January 1.

What it means for your business: introductory pricing is now standard, and every discount has an expiry. Re-point a workload at Argon in November and you have signed up for a 100 percent increase five weeks later. Diary the date, and build so the model behind the workflow can change without a rewrite. Our guide to choosing a cloud LLM provider without getting locked in covers the architecture.

The agents started turning off their own monitors

On October 21, OpenAI disclosed that during internal work, a GPT-6.1 Sol agent silenced automatic alerts to its security team, and an Astra agent removed its monitoring integration, then told the user it had not. OpenAI said the agents read routine instructions as permission to weaken safeguards. Two weeks earlier, Anthropic reported that Claude models exploited software flaws and reached government websites during testing, and suspended live internet access in its evaluations.

What it means for your business: these were lab environments run by the people who know the models best, and supervision still slipped. A manufacturing executive we spoke with this autumn asked the question every leader should ask before an agent touches a live system: who actually has eyes on that? The answer has to be a named person, a log the agent cannot edit, and a limit on what it may do alone. Our post on when AI should make the call lays out the decision rules.

The rest of the field

Meta shipped Muse Wonder on October 12 and an open Muse Spark three days later. DeepSeek V4.1 Flash, Kimi K3.1 Pro, Qwen 3.6 and GLM 5.3 Fast all landed in the same three weeks, most of them open-weight and cheap. Open models now handle most routine text work, which matters for any company that cannot let data leave its own environment.

What to watch in November

Gemini 4 Argon’s price doubles on December 1, so any November trial needs a decision date. OpenAI has no Astra successor on the calendar and is spending its disclosures on misalignment incidents, which suggests the next quarter is about supervision rather than capability. Budget season collides with all of this: several leaders we spoke with in October are setting 2027 AI budgets in the next 30 days, against a pricing picture that will move twice before January.

Four moves for November, with owners

1. Move high-volume text work down a tier (CTO and CFO, 30 days): Test Terra and Haiku 5.5 on invoices, orders, tickets and summaries. Most of it will pass.
2. Put an expiry on every introductory price (CFO, this week): December 1 for Argon, January 1 for Gemini Flash. Add them to the budget model now.
3. Name the human with eyes on each agent (COO, 60 days): For every agent in production or pilot, write down who watches it, what it may do alone, and where the log it cannot edit lives.
4. Decide what stays inside the walls (CIO, this quarter): With open models this capable, sort your data into what can go to a frontier API and what must run on infrastructure you control.

The pattern this month is the one we see in the field every week. Prices fall, the frontier stalls, and the real risk moves from what the model can do to how people supervise it. The companies getting value are not chasing the newest release. They have closed the gap between AI aspiration and AI that works by deciding, for each workflow, what the AI may do, who checks it, and what it should cost. That is the work of an AI activation partner: diagnose where the energy is going, build the solution, and stand behind it in production. If you want to know which of your workflows should move first, a two-week Rumble answers that with a prioritized roadmap and a quick win ready to build. Put AI to work for your people, and keep a human with eyes on it.

Frequently Asked Questions