Everyone wants to know which AI model is smartest when it comes to finance. TC Labs, Trading Central's innovation unit, put a different question to the test: what happens when you give the same model better market research data?

The answer came from FinMMEval 2026, an international AI evaluation lab run as part of CLEF 2026 in Jena, Germany. As part of the lab, our data scientists, Elvys Linhares Pontes and Mohamed Benjannet, entered a live trading challenge and found that adding high quality research data from Trading Central lifted the accuracy of every AI model they tested.

Here's what they did, what they found, and why it matters for brokerages planning their next AI move.

The test and the results

The challenge mirrors a real trading desk. Every day, each AI agent reviews the latest market news for Tesla (TSLA) and Bitcoin (BTC) and makes a call: buy, hold, or sell. It has three minutes to decide.

Elvys and Mohamed tested several AI models in three configurations to evaluate whether a smarter AI trader comes from better inputs or more training.

• Baseline Configuration: models are provided only with organizer-supplied data.
• Augmented Configuration: Trading Central news summaries and technical indicators are incorporated in addition to the organizer-provided data.
• Fine-Tuned Configuration: models are further trained on the provided training dataset.

The team found that adding news analysis and technical indicators from Trading Central helped the models better interpret market conditions and make better-informed decisions, leading to consistent improvements across all metrics. The top performer, DeepSeek-R1 with Trading Central data, outperformed buy-and-hold by 13.55% on TSLA and 15.57% on BTC in a falling market.

In contrast, fine-tuned versions of LLMs exhibited a reduction in performance, likely due to constrained reasoning capacity introduced during the fine-tuning process.

Read the full research at CEUR Workshop Proceedings.

What this means for brokerages

Most of the AI conversation in our industry centers on the model. Which LLM? How big? Should we fine-tune? Our results suggest a more practical question comes first: what is the model actually reading?

A general-purpose model knows a lot about markets in the abstract. It doesn't know where Tesla's support sits today, whether the price just crossed the 50-day moving average, or which headline actually moved the asset this morning. Without that, even a strong model is guessing with confidence.

For brokerages rolling out AI assistants, copilots, or AI-driven features, you don't need to win the model arms race to deliver smarter experiences. You need a trusted, structured intelligence layer your AI can draw on in real time, so your clients get answers grounded in the market as it is right now.

That intelligence layer is exactly what the Trading Central MCP server delivers. 

Through a single connection, your AI can access 59 tools across 8 analytical domains to deliver explainable, traceable financial insights backed by nearly 30 years of expertise and world-class research. Every data point is legally sourced from reputable data partners and rights-managed, allowing you to reduce compliance risk, bring your AI project to market faster, and empower your traders with more confidence.

Talk to our team to see how AI agents can produce smart, grounded market insights with Trading Central MCP.

‍