deepmarketmaking.com

What It Is

Deep MM’s core product, AXOR, is positioned as the financial market’s first Large Event Model. It is not a general-purpose large language model, nor a minor iteration on traditional quant models, but a credit bond pricing signal trained specifically on market events. Its current deployment focus is U.S. investment-grade and high-yield corporate bonds, with a clear target audience of systematic credit trading desks, market makers, and large financial institutions.

Core Capabilities

AXOR uses 80M parameters and extracts 2,000+ real-time features per inference. Data sources include FINRA TRACE, the U.S. Treasury curve, SPY, QQQ, HYG, JNK, LQD, and S&P ratings. Its key differentiator is that it does not provide only a single price point; instead, it outputs a 19-point probability distribution from the 5th to 95th percentile, conditioned by CUSIP, trade size, buy/sell direction, and ATS. This makes it better suited to RFQ pricing, risk boundaries, and large-trade decision-making. According to the official website, across 400,000+ out-of-sample trades, its IG/HY pricing accuracy improves by 44%/48% versus public benchmarks. It is also already in production at a Top-5 global bank, where it delivered a 12% MAE improvement as an ensemble input.

Integration and Pricing

The product is API-first and delivered in real time via WebSocket, refreshing every 15–30 seconds, with up to 32,000 subscriptions per connection. It requires no new terminal, and the company states that it does not need client data or require rebuilding existing models. The AXOR QX quoting engine includes sample code, consulting, and integration support, marked as free and under a perpetual license. However, pricing for the main product is not publicly disclosed; access requires booking a private demo, so procurement is likely closer to an enterprise-custom arrangement.

Pros, Cons, and Who It’s For

Its strengths are a clearly defined event-model approach, probability-distribution outputs that are useful for trading decisions, a relatively lightweight integration path, and real institutional production cases. Its limitations are that coverage is currently mainly U.S. IG/HY corporate bonds, methodology details must be requested separately, and pricing, SLA, compliance certifications, and Chinese-language support are not disclosed. It is not suitable for individual investors or typical AI tool users; it is better suited to credit trading institutions with internal quant, pricing, and execution systems.

China Access and Alternatives

There is no information in the main materials about access from China, so this is currently unknown; payment methods are also not disclosed. Chinese institutions evaluating it should focus on confirming network connectivity, cross-border data issues and market data licensing, USD payment, and contract compliance. Comparable alternatives or complements include MarketAxess CP+, Bloomberg BVAL, Tradeweb credit tools, and internally built corporate bond pricing models.