Noesis VC @NoesisVC

Sources: OpenAI, GPT-6.1 Terra announcement ($1 per million input tokens, $5 output; benchmarks are OpenAI's own). Moonshot, Kimi K3.1 Pro pricing (a third below K3).

Same work. Half the price. OpenAI's new GPT-6.1 Terra costs 50% less than the model it replaces, and OpenAI says it scores higher on professional computer tasks. Founders: build for next year's AI price. When it drops, be the team already shipping.

The test: write the reason your 20th hire joins in one sentence. If it could describe ten other companies, it isn't an answer yet.

Lockheed's Skunk Works had a rule for team size: restrict it "in an almost vicious manner." 10% to 25% of a normal programme. If your first build needs a big team, the scope is too big.

Quote via @StartupArchive_. Thiel's answer: "This is the only place in the world where you can work on this incredibly important problem."

Peter Thiel's hiring question doubles as fundraising prep for frontier founders. The first five people join for the founders and the equity. The 20th joins for the problem. Write your answer before you raise.

Sources: SiliconANGLE and Payload, Colare pre-seed announcement (Afore Capital, Garage Capital, Acadian Ventures). The 40-hour figure is Colare's estimate, via .

In deep tech, hiring is a bottleneck of its own. Colare raised a $2.5M pre-seed to test hardware engineers on simulated job tasks: CAD, circuit boards, thermal analysis. Its estimate: one mechanical, electrical or embedded hire can take about 40 hours of senior engineers' time.

Every frontier looks like a science project for exactly one round too long.

If your co-founder pitches it differently than you do, you don't have a pitch yet.

OpenAI asked its own agents to break out of its research sandboxes. They found flaws in almost every narrow internet connection left open: metrics, logging, authentication. Building agents? The sandbox is part of the product. Red-team it before a customer does. Source: OpenAI

What a pre-seed round in deep tech actually buys. ↓

Three rules for customer calls that don’t lie, from Rob Fitzpatrick’s The Mom Test: 1. Talk about their life, not your idea 2. Ask about specifics in the past, not opinions about the future 3. Talk less, listen more If they’ve never looked for a fix, they won’t buy yours.

Sources: Anthropic S-1 filed with the SEC (Nasdaq, ticker ANT, price range not yet set). Latest closed private rounds per Anthropic, OpenAI and Databricks. Revolut Ventures announcement, via Euronews and Europa Press. Via @maxkarpis

The AI boom was built in private. Anthropic filed to go public. It got to $965B in private rounds first. OpenAI ($852B) and Databricks ($190B) are still private. So Revolut plans to let eligible app users into private tech from 2027. Founders: plan for a long private life.

Frontier Signals: Minerva Humanoids came out of stealth with about $10M in pre-seed, led by General Catalyst. Idea to a walking humanoid in five months. Next: its first paid pilots. Robots for bomb disposal and hazardous energy work. Source: Minerva Humanoids announcement

Proof has a scale. NASA’s runs from 1 to 9, from a principle on paper to a system proven in operation. Know which level your round gets you to. ↓

If you can’t say why now, investors assume someone already tried.

Rounds: Enact, Stellios, Vessla, Delegance, Reduck, Zero Industries, Weflayr, RhymeQ, PataBid, HapWare, Micro Thermal Energy, Jenz, HeyVoli. Sources: company announcements, , BetaKit, FinSMEs, SeedCue, PR Newswire. Euro rounds at $1.16.

Last week's biggest pre-seed: $30M. The median: $2.5M. 13 AI and deep tech pre-seed rounds, side by side. One took more than half the money. 8 of 13 raised between $1.5M and $3.5M. Benchmark your round against the median, not the headline.