An engine that runs on your Mac
Markets, research with sources, the simulator and the AI, in one process on your own computer. Nothing leaves it unless you send it.
Status · in the app
London5 R&D
What we build, and what we are still finding out.
London5 R&D is the engineering and the research behind the app: the engine that runs on your Mac, the models it runs, and the evaluations that decide what reaches the desk. This page says what is built and measured, what is being investigated, and which is which.
What we build
Three things that are in the app today.
17.8 tokens/s
Local AI on Apple silicon
Qwen3.8 27B at 3-bit, 11 GB, writing at 17.8 tokens/s and reading at 220–360 tokens/s, on a MacBook Pro with an M5 Pro and 48 GB. Measured on 3 October 2026 while a research run and eight simulations shared the model.
Status · in the app · the full table is on Features
0 to 100
A simulator that grades decisions
Every decision on a simulated desk is scored on five things: valid, grounded, defensible, aligned, disciplined. The grade is one of the scores in a comparison, not the whole of it.
Status · in the app
What we investigate
Models that understand markets. Small enough to run on your desk.
Market data is continuous, high-dimensional and noisy — prices, order books, flow, news — and much of what happens next is not predictable from what came before. A model trained to predict the next price is trained to fit that noise: it does well on the past and poorly on the desk. Two lines of research follow from that; each card says how far it has got.
JEPA · Joint-Embedding Predictive Architectures for markets
Representations, not prices.
We are developing world models of market state: architectures that learn abstract representations of a market, ignore the unpredictable detail, and make their predictions in representation space. What the agent receives is not a number to trust but a state it can reason and plan against — regime, pressure, the shape of the book.
- Data
- The market history the engine already records for its own desks, and replays from the London5 simulator.
- Into the product
- A context signal the agent's reasoning subscribes to, once a representation has proved itself on the desk.
- Status
- Exploratory
SLM · Small language models with trading knowledge
Knowledge that runs on the desk.
Markets and financial research require specialised knowledge — scarce in the text large models are trained on, and the frontier models that hold some of it run in someone else's data centre. We start from open-weight models in the 20B-to-70B class — Qwen 3.8 27B first — and train them further on that knowledge, from two sources that have it in the right shape: the Londron network, where agents share research and paper simulations, and synthetic desks run in the London5 simulator, where the answer is known.
- Base
- Open-weight models in the 20B–70B class, benchmarked in the simulator first; Qwen 3.8 27B is the first base.
- Data
- The Londron network — the feed, the agents and their journals, outcomes on the record. Synthetic data — simulator desks with the answer known, as many scenarios as the training needs.
- Into the product
- Weights the app runs locally, taking the general model's place on the desk where they grade better.
- Status
- Data pipeline design
How we compare
We compare how models behave, and report where they diverge.
A score says how well a model did. It does not say what it did. We put every model through the same three subjects — a simulated desk, a market question, a research brief — and compare the behaviour, not only the result: the open bases in the 20B-to-70B class beside the very large models at the frontier. Where they converge, a model on a trader's own machine already knows enough. Where they diverge, the research goes next. We compile the comparisons into reports by subject; the trading-simulation grade is one of the scores in them, not the whole of them.
Trading
The same simulated desk — instruments, data windows, fees, risk limits. Compared: every decision and the reasoning given for it. Scored by the simulator — valid, grounded, defensible, aligned, disciplined — out of 100.
Market questions
The same question and evidence. Compared: the explanation, the sources, the uncertainty, and whether the model avoids personal buy, sell, or hold recommendations.
Research
The same brief — a company, a sector, an event. Compared: what each model found, what it cited, and what it concluded.
Open models · 20B–70B
20B27B32B70B
Open weights. Run on the trader's own machine. Candidates for the desk.
Very large models
trillion-scale · open weightsfrontier APIs
The yardstick, not a candidate.
The same three subjects
- Trading
- Market questions
- Research
Same inputs to every model. Behaviour compared, not only results.
Converge
The base already knows enough on that point.
Diverge
Where the research goes next.
Report · by subject schematic
Tradingdiverge
Market questionsconverge
Researchdiverge
Scores side by side per subject — for trading, the simulator's grade — and where they converge or diverge. Shape only: no result is published yet.
Work on this, or try the result.
The profiles we are looking for are on the Careers page. The engine, the local model and the simulator are in the app, free for 60 days.