Thinking Machines put Inkling 2 Small, a 594-billion-parameter mixture-of-experts model, on Tinker with Apache 2.0 weights planned. Agents also move into video calls and checkout flows.
Koda editorial team · 14 November 2026 · 6 min read
Listen · 16 min
Today: Test one new model and one new agent habit The steps
On 5 November, Thinking Machines launched Inkling 2 Small, a 594-billion-parameter mixture-of-experts model with 17 billion1 active parameters, pretrained in 68 hours on NVIDIA GB300 systems. It scores 31 on the Artificial Analysis Intelligence Index and ships through Tinker, with Apache 2.0 weights planned. On 13 November, DeepSeek opened its app features to developers through its API.
Why it matters. Because only 17 billion1 parameters run per token, Inkling 2 Small should cost about as much to run as a far smaller model.
Go further: today's paper, DeepFinance: Can AI Diligence a 100MM Deal? (arXiv, 7 Nov).
OpenAI launched ChatGPT Wallet so agents can handle transactions and manage finances, as Replit Agent gains purchasing for businesses.
Why it matters. A wallet inside ChatGPT lets agents pay without a human at checkout, so finance teams need per-agent spending limits and audit logs before ChatGPT and Replit Agent buying spreads.
Gemini 4 Hydrogen entered the thinQit Index at 19th this week with 71.1. Gemini 4 Carbon leads at 83.2, ahead of Claude Opus 5.5 at 82.0 and Claude Fable 5.2 at 81.5.
Why it matters. Gemini 4 Hydrogen trails Gemini 4 Carbon by a wide margin.
n8n released Nod, an Apache 2.0 automation agent that runs locally, as Anthropic moves Claude Managed Agents inside security perimeters.
Why it matters. Local execution keeps workflow data on the device, so privacy review shrinks to one machine; Claude Managed Agents reaches a similar goal through Anthropic's hosted models.
Anthropic's Claude Meetings lets teams add Claude to video calls, where it follows along, takes notes and starts on follow-up work before the call ends. It is rolling out now to Team and Enterprise plans on Google Meet, with other platforms coming later. The launch lands days before Anthropic's IPO.
Microsoft's global outage disrupted Outlook, Teams features, SharePoint and Copilot integrations for businesses, governments, airlines and banks on Friday, November 13. The failure sits in Exchange Online and the features built on it. Recovery has been uneven: servers restored overnight failed again.
The analysis argues that a well-designed B2B AI agent can double annual contract value by performing work humans used to do, which lets vendors draw on labor budgets beyond limited software budgets. Most AI features only improve existing workflows and leave ACV flat.
Iran's Guards had claimed no strike on U.S. assets since early September; now they say they hit a U.S. radar in Bahrain. CENTCOM says it downed two drones and the radar survived.
Why it matters. The first claimed strike on U.S. assets since early September suggests Iran's Guards are escalating again.
The Nasdaq fell 0.91% on the 13 Nov session as the S&P 500 slipped 0.40% and WTI crude edged up 0.20%, so the selling that day stayed in equities; separately, Bitcoin dropped 5.15% on 14 Nov with the crypto Fear & Greed index at 13, Extreme Fear.
S&P 500
7,561.67
-0.40%
Nasdaq
25,136.27
-0.91%
Bitcoin
$67,666.18
-5.15%
Ethereum
$1,910.15
-5.29%
Crude Oil (WTI)
$88.92
+0.20%
Crypto Fear & Greed
13 / 100
Extreme Fear
Levels from Yahoo Finance2, 13 Nov and 14 Nov; alternative.me3, 14 Nov.
uKit AI sells a migration as much as a makeover: give it your old URL and it rebuilds the site inside uKit's own builder with the content carried across, in about 10 minutes, uKit says. Owners of plain brochure or portfolio sites lose little by trying a preview, and the bet is worth taking if the result beats their current theme. Anyone running a big catalog or relying on hand-tuned phone layouts should skip it.
Pick one workflow and measure it this week.
A lab that can pretrain a 594-billion-parameter model in 68 hours, while Claude Meetings and ChatGPT Wallet move agents into calls and checkout, means the scarce resource is now your own evaluation data. Choose a single workflow, such as meeting follow-ups or a purchasing step, and score two or three options on it with your own prompts. Keep a written record of the failures, since that record is what will let you switch models quickly once open weights arrive.
Every story links to its outlet. Below, by source, are the 8 figures we checked: 0 verified, 7 reported and 1 estimated.
Verified confirmed by an independent searchReported as published by the named sourceEstimated approximate, in the source or in our check
Signal 232 · 14 November 2026 · 15 sources · How every number is checked · Corrections
231 editions so far, one a day. Unsubscribe anytime.
Forward this to one operator you work with. Your referral link is in every email; milestones at koda.community/refer.