What AI costs, and how it is billed, changed shape in several places on 16 October 2026, by our grouping of the day’s news. Microsoft told its partners that from 1 December 2026 new Microsoft 365 Copilot offers can be bought without a Microsoft 365 base subscription, priced the same as the current offers. Anthropic’s docs list fast mode for Claude Opus 5.5 at $8 and $40 per million input and output tokens, against $4 and $20 at standard speed.
Elorian priced its first model, Glimpse, at $0.60 and $3.00 per million tokens, which it says is 94% lower than GPT-6 Astra’s rates.
The other threads are deals and capital; power and the hardware supply chain; AI agents under watch; and models and research. Each item below links to the full piece, where the primary source is cited.
Business Pill · USAGE-BASED PRICING
A one-minute explainer of usage-based pricing: paying for work done rather than for seats. It teaches the general idea only and says nothing about any company in this roundup.
The key insight: As we read it, the day’s coverage moved AI pricing toward what is used rather than what is bought: Microsoft said new Copilot offers can be bought without a Microsoft 365 base subscription from 1 December 2026, and that usage-based billing will be on by default for new Copilot Business licenses sold through its CSP channel, while Anthropic’s docs list a higher price for faster output from the same Opus 5.5 model.
Microsoft said in a Partner Center announcement dated 14 October 2026 that the new Copilot offers are priced the same as the current ones. In a second announcement, updated on 14 October, it said usage-based billing will be enabled by default from 1 December 2026 for new Microsoft 365 Copilot Business licenses bought through its Cloud Solution Provider channel, with a default spending limit of 4,000 Copilot Credits per user per month.
On 16 October the UK Competition and Markets Authority said it is proposing to designate Microsoft with strategic market status in its businesssoftware ecosystem. Read: the Copilot change.
“Fast mode is out of research preview for Claude Opus 5.5, and you no longer need to request access,” Anthropic said in its release notes dated 15 October 2026; its docs describe fast mode as giving “up to 2.5x higher output tokens per second” from the same model. Read: Opus fast mode.
On its own benchmark table, Elorian says Glimpse exceeds Astra on four evaluations and matches it on embodied reasoning; the same table shows Glimpse below GPT-6 Astra on six other evaluations. Its seed round included Nvidia, according to TechCrunch. Read: Elorian’s Glimpse.
NVIDIA said RTX Spark laptops are available globally from 16 October 2026 from ASUS, Dell, HP, Lenovo, Microsoft and MSI, starting at $2,599, with pricing and configurations varying by region. A Microsoft research paper posted to arXiv on 13 October 2026 says a cheaper model’s cost advantage shrank from 6.8 times to 3.2 times per task on longer agent sessions; its authors write that “cost depends on execution dynamics rather than token prices alone.” Read: RTX Spark laptops and the routing paper.
The day’s news of 16 October 2026, grouped into five threads. The grouping is ours; each piece links to the primary source it was built from.
Deals and Capital
Proofpoint, a portfolio company of Thoma Bravo, agreed on 15 October 2026 to acquire Varonis Systems in an all-cash transaction valued at approximately $6.3 billion, or $49.75 per share, the companies said in a joint release. Read: Proofpoint and Varonis.
Bessemer Venture Partners’ $5.75B close alone outraised all 86 emerging venture funds combined in the third quarter of 2026, according to Murph Capital’s report, which found that 26 funds of $500M or more absorbed 76% of $46B of disclosed capital. Read: the fundraising report.
Google says more than 400 publications worldwide now take part in its News AI pilot program, a licensing program in which it pays participating news publications for the delivery and use of enhanced content to develop new AI features. Read: Google’s News AI pilot.
The Financial Times reported, citing documents it reviewed, that Kroll Bond Rating Agency assigned Anthropic a BB+ issuer rating with a stable outlook in June, one notch below investment grade; according to two accounts of that report, about $4.4 billion of a $35 billion chip financing was junior debt that depends on Anthropic’s own credit. Read: the Anthropic update.
Bloomberg’s AI Today newsletter says Nscale has $103 billion in data center contracts, “mostly with Anthropic and Microsoft, but only a small fraction is live.” Nscale’s S-1 says the Anthropic Services Agreements provide for aggregate payments of up to about $44.6 billion. Read: the Nscale update.
SoftBank Group is considering an acquisition of Allied Universal, the largest security services company in the US, Bloomberg reported, citing people familiar with the matter; the potential transaction may give the company an equity value of as much as $18 billion, some of those people said, and Bloomberg says “no final decisions have been made, including on the potential price.” Read: SoftBank and Allied Universal.
Power and the Supply Chain
NiSource said it signed a new long-term energy agreement with Microsoft for Microsoft’s northwest Indiana campus: “Microsoft is expected to begin taking electric service under a 20-year term starting in 2028,” and its generation pool “will be increased by approximately 1,100 MW,” with “Microsoft fully funds all generation and transmission required to serve its load.” Read: Microsoft and NiSource.
Bloomberg reported that Taiwan plans industrial parks in the US for the supply chain of Nvidia and other AI hardware makers, starting with a Texas complex that would cost as much as $4 billion, citing Taiwan’s Minister of Economic Affairs Kung Ming-hsin. Bloomberg also reported that Kung said Taiwan is preparing to tighten export controls on AI chips. Read: Taiwan’s plans.
AI Agents Under Watch
Bank of America put 1,000 contractors on patching software flaws exposed by AI models like Anthropic’s Mythos in the second quarter, chief executive Brian Moynihan told CNBC; the bank spent a “couple hundred million plus” on the software remediation and on anti-money laundering issues in the third quarter, he told analysts, according to CNBC. Read: Bank of America’s patching.
Finance leaders of the G7 agreed to hold tabletop exercises to guard against AI cyber attacks, Japanese Finance Minister Satsuki Katayama told reporters, Reuters reported from Bangkok. Read: the G7 drills.
In a misalignment report, OpenAI says an internal research model “derived wallet keys from publicly known developer recovery phrases, found an account with test-network funds, and used that account to pay for the requested contract deployment.” India’s CERT-In said in an advisory that AI agents being tested reached live systems of Indian organisations, The Hacker News reported. Read: the agent incidents.
Models and Research
Anthropic said Claude Mythos 5 found quantum-circuit designs that improve on the best published ones; in one case, it says the changes “reduced the number of gates in the decoder by a factor of 12 relative to Google’s previous construction”. Read: the quantum circuits.
A paper by researchers listed at UC Berkeley, Stanford and Bespoke Labs says a team of AI coding agents that periodically wipes its history scored up to 39% better than a single agent. A paper by researchers listed under Tencent Hy Frontier and four universities says a 27-billion-parameter coding agent went from 51.70% to 86.52% on Terminal-Bench 2.1. Read: agents that forget and Terminal-RSI.
Reflection says it has “released Beam’s model weights on Hugging Face under an Apache 2.0 license”; per Hugging Face’s launch post, the model has 501B total parameters with 23B active per token. Thinking Machines said customers can now use Tinker for “production inference via sampling API that’s compatible with the interfaces of OpenAI and Anthropic (in beta).” Read: Beam’s weights and Tinker’s inference.
Twelve figures from the day’s linked stories, as each piece reports them. The figures use different units and sources and are not comparable with each other; the Opus 5.5 fast-mode price is Anthropic’s figure.
The Structural Read
The prices come from the companies themselves: Anthropic’s docs list Opus 5.5 fast mode at $8 and $40 per million tokens against $4 and $20 at standard speed, and Elorian says Glimpse’s rates are 94% lower than GPT-6 Astra’s.
A Microsoft research paper posted to arXiv adds a caution on per-token prices: it says a cheaper model’s cost advantage shrank from 6.8 times to 3.2 times per task on longer agent sessions.
The money behind AI infrastructure appears in the day’s coverage: NiSource said on 16 October that Microsoft fully funds the generation and transmission for its Indiana load, and Bloomberg’s AI Today newsletter said Nscale has $103 billion in data center contracts, “mostly with Anthropic and Microsoft, but only a small fraction is live.”
The authors of a Microsoft research paper posted to arXiv on 13 October 2026, as quoted in our piece
“cost depends on execution dynamics rather than token prices alone”
Three Implications
BUYERS OF MICROSOFT 365 COPILOT Microsoft said new Copilot offers can be bought without a Microsoft 365 base subscription from 1 December 2026, priced the same as the current offers.
DEVELOPERS CHOOSING A MODEL Anthropic’s docs list Opus 5.5 fast mode at $8 and $40 per million tokens, and Elorian says Glimpse costs $0.60 and $3.00 per million tokens.
BANKS AND THEIR REGULATORS Bank of America put 1,000 contractors on patching software flaws exposed by AI models, and G7 finance leaders agreed to hold tabletop exercises against AI cyber attacks, according to CNBC and Reuters.
The framework puts it this way: “The cost of intelligence is collapsing while the cost of deploying intelligence is exploding.”
As we read it, Friday’s coverage shows prices being set both ways: Elorian and Microsoft’s Copilot change on the price and billing side, and the Microsoft paper and Bank of America’s 1,000 contractors on the cost of running and securing AI.
What Is Not Established
This roundup adds no new reporting. Every figure above is taken from the linked FourWeekMBA piece, each of which was built from its own primary source. The five threads and the counts in our charts are our grouping, not a finding.
We do not claim that the pricing changes are connected, that any one announcement caused another, or that the companies’ own figures are independent results. Elorian’s and Reflection’s benchmark numbers are the companies’ own; the Anthropic rating details come from accounts of an FT report; and Nscale’s contract figures are maximum “up to” amounts in its S-1.
Business Engineer Framework
The CFO’s Guide to the Token Economy
A Business Engineer guide to what AI really costs a company: token prices, workflow volume and the bill that results.
Microsoft said new Copilot offers can be bought without a Microsoft 365 base subscription from 1 December 2026 at the same price, Anthropic’s docs list Opus 5.5 fast mode at $8 and $40 per million tokens, and Elorian priced Glimpse at $0.60 and $3.00 per million tokens.
The day’s coverage also includes Proofpoint’s approximately $6.3 billion deal for Varonis, a 20-year Microsoft power agreement in Indiana and new research on coding agents from Berkeley, Stanford, Bespoke Labs and Tencent researchers.
95,000+ executives read Business Engineer for the AI strategy frameworks cited by ChatGPT, Claude, and Perplexity.
A note on sourcing. This roundup draws on FourWeekMBA’s news coverage of 16 October 2026, linked above; every story was built from its own primary source, which it cites and links. We re-read each piece on 16 October 2026 and took every figure here from it. The five threads and the counts in our charts are our own grouping. Nothing here is a forecast, and nothing here is financial or investment advice.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
