BMLL, an independent supplier of harmonised historical Level 3, 2 and 1 data and analytics for capital markets, has joined forces with Simudyne, a generative AI and agent-based simulation platform, to let market participants run realistic and reactive simulations of trading conditions.
Under the deal, BMLL’s harmonised historical Level 2 and Level 3 order book datasets will be paired with Simudyne’s Pulse simulator, which is built for intraday markets. Firms will be able to carry out market replays, assess how their systems perform and run simulations that react to their own activity.
Simudyne has started using BMLL’s granular data to train what it calls generative AI “Large Market Models”. These models support the creation of synthetic tick data and can be fine-tuned for individual clients. They sit alongside the company’s existing rule-based agent-based modelling tools. Unlike a fixed replay of past trading, the simulator supports reactive back-testing, in which the simulated market shifts in real time in response to orders placed by the user.
The combined offering is aimed at several groups across capital markets. Algorithmic traders and quants can refine execution strategies against a market that responds to them, and can use Simudyne’s calibrated models and BMLL’s data as a benchmark for training their own machine learning execution models.
Execution and TCA analysts can model large orders before placing them to gauge slippage, costs and market impact, helping them choose routing strategies. Trading technology teams can link staging and test systems to the simulated exchange through FIX connectivity or APIs to check latency and overall performance.
Exchanges can trial changes to matching engine logic, protocols and new order types on a replica before going live, allowing them to prototype auction mechanics, circuit breakers and fee structures without risk. Risk and clearing teams, meanwhile, can carry out complex liquidity risk calculations and stress-test margin, clearing and default-fund models across a variety of simulated volatility scenarios and market regimes.
Simudyne develops market simulation software using techniques including agent-based modelling, which draws on proprietary behavioural assumptions about market participants, as well as other machine learning approaches. The firm is now pre-training generative models of intraday order flow using methods similar to those behind language models such as ChatGPT. Because these models are not based on hand-written expert rules, how realistic they are depends entirely on the volume and quality of the data used to train them.
The tie-up stems from Simudyne joining the BMLL Activate Data Credits Programme. The scheme supports partners building, testing and launching data-driven products and analytics on BMLL’s historical Level 3, 2 and 1 market data. It reduces or removes upfront data licence fees during development, and chosen partners are given a set number of data credits that can be exchanged for structured access to the BMLL Data Lab and BMLL Data Feed. This will allow Simudyne to build, train and host its generative foundation models and multi-agent simulation software, giving joint clients a route from research and validation through to commercial use.
BMLL CEO Paul Humphrey said, “We are delighted to welcome Simudyne to our BMLL Activate Data Credits Programme. High-quality training data is the single most critical ingredient for sophisticated, AI-driven financial modelling. By hosting Simudyne’s simulation models directly within the BMLL Data Lab, we will be able to give our clients a safe, highly realistic sandbox to run reactive simulations on top of our definitive historical order book.
“Ultimately, this collaboration serves as a powerful proof-of-concept, establishing BMLL data as the premier foundation for firms to train their own AI models, while reinforcing our ongoing commitment to driving the next wave of innovation across global Capital Markets.”
Simudyne CEO Justin Lyon said, “We are very excited to collaborate with BMLL to push the boundaries of market simulation. As we move towards data-driven, generative AI models, simulation fidelity is only as strong as the data underpinning them. BMLL’s granular, high-resolution historical market data gives our models the depth needed to learn and reproduce real market dynamics at the order-book level. Together, we are creating the next generation of market simulation environment that bridges historical market behaviour and realistic synthetic markets.”
Copyright © 2026 FinTech Global
