Mercado Weekly summary November 2026
What changed in AI and payments in the week of November 9 to 15, 2026
22 notes
- 01
Generator queue slots become a strategic AI infrastructure asset
According to the Financial Times, reported on November 8, 2026, Anthropic agreed to pay about $3.4 billion for 1.7 GW of gas-fired generators; the company has not confirmed the deal.
Read the note → - 02
With the model fixed, the harness shows up in cost and speed
In an open evaluation dated November 4, 2026, Garry Tan held the model and tasks fixed and swapped only the harness; pass rates were statistically indistinguishable, while cost per solved task diverged.
Read the note → - 03
Rivio closes Series A with agents on the hospital revenue cycle
R$180 million in a Series A extension and a valuation above R$1 billion, according to a Rivio release dated November 4, 2026. The model combines AI agents, healthcare specialists and fees tied to results.
Read the note → - 04
BMO puts an agent on card inquiries and aims to reuse the platform
BMO has put an agentic assistant into card servicing at Virtual Connect, handling matching, policy and case prep before a human decides. The time and automation figures released on November 6, 2026 are the bank's targets.
Read the note → - 05
PNC starts accepting USDC for corporate clients and pays out in dollars
On November 9, 2026, PNC Bank launched a pilot that lets corporate and institutional clients receive USDC payments, automatically converted 1:1 into dollars through Coinbase and deposited into a PNC account.
Read the note → - 06
Google builds TPU 8tx to run AI on sensitive data with hardware protection
On November 9, 2026, Google announced TPU 8tx, an eighth generation TPU system for secure and confidential inference that keeps data and model weights protected in hardware during processing too. Availability dates and pricing were not disclosed.
Read the note → - 07
Meta launches Muse Ember for the repetitive work of agents
On November 9, 2026, Meta released Muse Ember, its smallest and most efficient model, at $0.10 per million input tokens and $0.50 per million output. Artificial Analysis puts it level with Claude Haiku 5.5 on its Intelligence Index at about a third of the cost per task.
Read the note → - 08
Anthropic gates Fable and Mythos 5.2 by verified purpose
On November 9, 2026, Anthropic released Claude Fable 5.2 and Claude Mythos 5.2, one model with two levels of access, priced 20% below the 5.1 generation and with outputs up to 30% faster. The physics results it cited are preprints that have not been peer reviewed.
Read the note → - 09
OpenAI opens business profiles and agents inside ChatGPT
OpenAI opened Businesses in ChatGPT in beta. A company can claim its profile, keep product feeds with pricing and availability up to date, and see mentions, clicks and discovery topics. The Business Agent, for now, is limited to eligible ecommerce and online service businesses in the United States.
Read the note → - 10
Gemini 4 Carbon arrives with a focus on financial research
Google DeepMind released Gemini 4 Carbon for long, complex work, the kind with several steps and a long verification loop. The finance results, which the company describes as pre-release evaluations, show Carbon matching the best model tested on Hebbia's financial agent benchmark and ranking first out of 23 setups on Balyasny Asset Management's financial intelligence benchmark. The introductory API price is $2 per million input tokens and $10 per million output, the same as Argon, before it moves to $4 and $20.
Read the note → - 11
Mercor announces a $750 million Series D to evaluate and train agents on each company's work
Mercor announced a $750 million Series D at a $20 billion valuation, twice the previous $10 billion mark. The round includes an employee tender. The plan is to bring enterprises the same custom evaluation and post-training infrastructure the company built for frontier labs, with a network of more than five million experts.
Read the note → - 12
IBM and Microsoft put agents, data and governance together in Imagine Enterprise
IBM and Microsoft announced Imagine Enterprise, a platform for companies to build, run and govern AI agents wired into the organization's own data and processes. It starts with financial services, insurance and retail. IBM Consulting brings more than 6,000 ready-made agents and assets. SEI, a technology and operations provider for the financial industry, appears as an example in an exploration phase, not as a case already in production.
Read the note → - 13
Google launches an API for fast decisions, with a confidence score
Google launched the Gemini Decisions API, powered by Gemini Decision 1. The call returns probabilities and confidence for up to 1,000 closed questions in a single pass, in 150 to 250 milliseconds, at $0.10 per million input tokens, with no output tokens billed.
Read the note → - 14
Fractile raises $720 million for inference chips it still plans to make
British firm Fractile announced a $720 million Series C, led by Thrive Capital, Founders Fund, Lightspeed and Redpoint, with the Sovereign AI fund and the British Business Bank. The UK government put the valuation at over £5 billion. The company plans to ship its first chips and systems next year.
Read the note → - 15
Runway opens the weights of Praxis-1, a model that predicts video and action together
Runway released Praxis-1 with open weights under an Apache 2.0 license, including for commercial use. It is a 27 billion parameter mixture of experts diffusion transformer that predicts video and actions together. Partners including Noble Machines, Standard Bots and Ultra were already running it on their own hardware.
Read the note → - 16
Torus launches Foundry and raises $500 million for inference at sites already on the grid
Torus launched Foundry and announced $500 million in growth capital, led by Magnetar, with ICONIQ, Origin Ventures, Ridgeline Partners and Trinity Capital. Modules run from 1 to 20 MW per site, behind the meter, with 12 to 16 weeks from contract to a running workload. The first named customer is Hydra Host.
Read the note → - 17
AWS Emulator enters general availability, written by agents, Amazon says
AWS launched Emulator in general availability, with no extra service fee, and the source on GitHub under CC0 1.0. Amazon says thousands of autonomous agents wrote every line, with a merge gate and a security agent in the review.
Read the note → - 18
OpenAI launches ChatGPT Wallet, with a network token and a check on every purchase
OpenAI launched ChatGPT Wallet, a way for an agent to pay without the business or the agent receiving the actual card details. After approval, the network issues a payment token. Every purchase needs a security check, such as Face ID or a fingerprint. It starts in dots, for ChatGPT Pro subscribers in the US who are 18 or older and have eligible US-issued cards.
Read the note → - 19
LangChain raises $300 million in a Series C, at a $3 billion valuation
LangChain announced a $300 million Series C at a $3 billion valuation, led by existing investor CapitalG, with significant participation from GIC and DFJ Growth and from existing investors IVP and Sapphire Ventures. The company says more than 7,000 organizations already rely on LangSmith. In the official post, LangChain reports that Workday went from five months to under a day to put a new agent live.
Read the note → - 20
Sierra releases Poppy 1.0, already on Shopify stores and in Mercury accounts
Sierra published Poppy 1.0. Every Shopify store publishes a Poppy file. Muse, Meta's agent, looks for Poppy first on every site. At Mercury, balance checks are live; a transfer still goes through the approval policy.
Read the note → - 21
Google adds Google Wallet to Gemini, for personal finance insights in the U.S.
Google Wallet joins Personal Intelligence on Gemini in the U.S. Spending habits, charts and subscriptions, with Plaid. Gemini can only read; it cannot move, manage or invest the money. You can turn it off.
Read the note → - 22
Deepnote releases the Agent Initiative Benchmark, with cost per task and the failure to stay quiet
Deepnote published the Agent Initiative Benchmark: 12 models, 50 situations, 180-second budget. GPT-6.1 Sol scored 8.65 at $0.11 per task. Overall, when silence was right, models stayed quiet only 6 times out of 10.
Read the note →