STOCK TITAN

Fr8Tech AI Lab Publishes Two Studies on AI Negotiation and Voice Agents

(Very High)

Sentiment and the balance of points

Rhea-AI Sentiment reads the wording of the document, how positive or negative its language is on a 1 to 5 scale. The balance of points shown with the takes weighs what the document actually discloses, so the two can disagree, for example when a trial that missed its main goal is described in upbeat language.

Tags
AI

Freight Technologies (FRGT) published two AI Lab studies on automated freight-rate negotiation and Spanish-language voice agents, and released an open-source toolkit for testing negotiation agents.

In a simulated freight market, the two best purpose-built engines outperformed the tested language models in 32 of 36 comparisons and never lost. One engine’s deal-closing rate fell from 62% of trading days in a balanced market to 16% when demand was scarce. In 231 simulated calls in Spanish, the direct speech-to-speech system misinterpreted less than 3% of spoken amounts, versus approximately 13% for the two systems that used transcription. The studies inform development of the Zayren platform and voice features.

Loading...
Loading translation...

Positive

  • Minor pointTwo purpose-built negotiation engines beat nine tested language models in 32 of 36 statistically corrected comparisons, with no losses.
  • Minor pointDirect speech-to-speech processing misinterpreted less than 3% of spoken monetary amounts versus approximately 13% for the two transcription-based systems.
  • Minor pointThe direct speech-to-speech system alone avoided recording agreements the simulated carrier had not accepted.
  • Minor pointOpen-source negotiation toolkit lets researchers and logistics companies benchmark their agents under Fr8Tech’s test protocol.

Negative

  • Minor pointOne reinforcement-learning engine closed deals on 16% of simulated trading days under scarce demand, down from 62% under balanced conditions.

News Explained

The release says that, alongside research informing ongoing product development, components from the seven capability areas described in the AI Lab section are already running in Fr8Tech products.

Key Figures

Winning comparisons: 32 of 36 comparisons Deal-closing days: 62% Deal-closing days: 16% +3 more
Winning comparisons
32 of 36 comparisons
Purpose-built engines versus the tested language models
Deal-closing days
62%
Reinforcement-learning engine in a balanced market
Deal-closing days
16%
Same engine competing for scarce demand
Recorded calls
231 calls
Spanish voice-agent study
Misinterpreted monetary amounts
Less than 3%
Speech-to-speech architecture in the Spanish voice study
Misinterpreted monetary amounts
Approximately 13%
Two cascaded pipelines in the Spanish voice study

Previous AI Reports

1 past event · Latest: Oct 27
Same Type 1 event
  1. Oct 27

    AI platform expansion

    24h Move
    +16.0%

    Expanded Zayren with simultaneous carrier negotiation and analytics tools for logistics teams.

24h Move is the share-price change in the day after each event; other market factors may also have contributed.

Key Terms

reinforcement-learning, large language models, cascaded pipelines, speech-to-speech architecture
4 terms
reinforcement-learning technical
"a reinforcement-learning engine that closed deals"
A machine-learning method where a computer learns to make decisions by trying actions and receiving feedback on results, like training a dog with treats or improving performance by playing a video game and learning what wins. Investors care because reinforcement-learning can power automated trading, pricing, and operational systems that adapt over time to changing markets, potentially boosting returns or efficiency but also creating model risk and hard-to-predict behavior.
large language models technical
"nine open-weight large language models (LLMs)"
Large language models are advanced AI systems trained on vast amounts of text to understand and generate human-like writing, like a very fast reader and writer that learns patterns in words and sentences. They matter to investors because they can change how companies operate—automating customer service, speeding analysis, cutting costs, creating new products—and they introduce risks around accuracy, security and regulation that can affect a firm’s revenue and reputation.
cascaded pipelines technical
"Two were cascaded pipelines"
A sequence of linked processes or workflows in which the output from one stage becomes the input for the next, so work flows downstream through multiple, dependent stages. Cascaded pipelines are used in manufacturing, chemical processing, software/data engineering and similar settings to break a complex task into ordered steps; a change or failure in an upstream stage affects all downstream stages, and throughput is constrained by the slowest stage in the chain.
speech-to-speech architecture technical
"a native speech-to-speech architecture"
A system design that takes spoken audio as input and produces spoken audio as output without requiring a human to transcribe to text and re‑read it; typical pipelines combine automatic speech recognition, processing (for content, translation or voice attributes) and text‑to‑speech or use end‑to‑end neural models that map input waveform to output waveform. It differs from speech‑to‑text or text‑to‑speech alone by covering the full audio‑to‑audio flow (often including language translation or voice conversion) and is optimized for low latency and naturalness when preserving meaning, timing or speaker characteristics.

AI-generated analysis. How Rhea-AI works. Not financial advice.

See more from StockTitan in Google Search and AI answers. Adds StockTitan as a preferred source · opens Google
Add on Google

Two studies test how AI agents negotiate freight rates under competition and how voice agents perform in Spanish; the Company also releases an open-source research toolkit for benchmarking negotiation agents and launches an AI Lab section on its redesigned website

HOUSTON, Nov. 20, 2026 (GLOBE NEWSWIRE) -- Freight Technologies, Inc. (Nasdaq: FRGT, "Fr8Tech" or the "Company"), a technology-centric logistics company offering a diversified portfolio of AI software solutions designed to address key inefficiencies in the supply chain, today announced that its in-house AI Lab has published two research papers on the design of AI agents for freight negotiation, together with an open-source research toolkit for benchmarking negotiation agents. The papers are available as preprints on arXiv and in a new AI Lab section of the Company's redesigned website.

The first paper informs the negotiation agents of Zayren, the Company's AI-based pricing-prediction, carrier-matching and negotiation platform. The second paper examines the voice architecture the Company is building into its products. The AI Lab ran controlled experiments to measure how AI architectures behave in freight-specific conditions and used the results to inform the ongoing development of Fr8Tech's products. The AI Lab section of the website details seven capability areas, from pricing and AI agents to computer vision and document intelligence, with components from these areas already running inside the Company's products.

The first paper, "Before AI Agents Negotiate for Us" (arXiv:2611.15838), compares nine negotiation engines, from rule-based and optimization designs to reinforcement-learning policies, run simultaneously on both sides of a simulated freight brokering market, under balanced conditions and under shortages of carriers or shippers. Through the same protocol, nine open-weight large language models (LLMs), ranging from 1 billion to 72 billion parameters, were run on the Lab's own hardware. In the end, the two strongest purpose-built engines outperformed the nine LLMs tested in 32 of 36 statistically corrected comparisons and never lost. The testing also exposed a failure: a reinforcement-learning engine that closed deals on 62% of trading days in a balanced market dropped to 16% when it had to compete for scarce demand. Catching failures like this in simulation, before an agent reaches a customer, is the purpose of the Lab's testing. The Company believes these findings support its approach of building negotiation agents on purpose-built architectures tested under simulated market stress, rather than just on a general-purpose language model.

"Our AI Lab exists to ground our technology in evidence-based research, not on assumptions about what it can do," said Javier Selgas, Chief Executive Officer of Freight Technologies. "Placing AI upstream of traditional transportation management means letting software negotiate, monitor, and execute. But before we hand those decisions to an agent, we want to know, with data, how it will behave when the market turns against it. Publishing this work openly is part of building that trust with our customers and with the market."

The second paper, "Latency, Speech Recognition and Turn-Taking in Voice Agents for Spanish Phone Negotiation: An Externally Measured Comparison of Pipecat, LiveKit and GPT-Live" (arXiv:2611.19118), evaluates three voice-agent architectures tested across 231 recorded calls in Spanish, simulating freight negotiations with a scripted carrier. Two were cascaded pipelines, in which separate components transcribe the carrier's speech, generate the reply and synthesize the voice; the third was a native speech-to-speech architecture that listens and speaks directly, without intermediate transcription. In the study, the speech-to-speech architecture misinterpreted less than 3% of spoken monetary amounts, compared with approximately 13% for the two pipelines, and it was the only one to not record an agreement that the carrier did not accept. These results inform the voice capabilities the Company is building into its products, from voice-driven negotiation to voice control of its software.

"We do not assume what an architecture can do; we build the test environment, run the experiment and let the result decide," said Umberto León Domínguez, Director of Fr8Tech's AI Lab. "In the negotiation study, no language model we tested won a single comparison against the strongest purpose-built engines. In the voice study, the difference between architectures was in how they listen, not in how fast they talk. Both answers came from measurement, not from assumptions, and both are shaping what we build."

The research toolkit is available on GitHub under an Apache 2.0 license, so that researchers and logistics companies can benchmark their own negotiation agents under the same protocol.

To read the papers and learn more about the AI Lab, please visit fr8technologies.com/en/ai-lab

About Freight Technologies Inc.

Freight Technologies, Inc. (Nasdaq: FRGT) ("Fr8Tech") is a technology company offering a diverse portfolio of platform solutions that are powered by AI and machine learning to optimize and automate the supply chain process. Focused on addressing the distinct challenges within the supply chain ecosystem, the Company's portfolio of solutions includes the Fr8App platform for over-the-road (OTR) B2B cross-border shipping across the USMCA region; Fleet Rocket, a nimble, scalable and cost-effective Transportation Management System (TMS) for brokers, shippers, and other logistics operators; Fr8Radar, a GPS integration and shipment visibility module within Fleet Rocket that enables real-time tracking of in-transit freight across the U.S.-Mexico corridor; Zayren, an AI-based pricing-prediction, carrier-matching and negotiation platform for OTR freight shipments across North America; and DODA Smart, an AI-powered customs compliance platform to automatically monitor, verify and trace Digital Customs Documents (DODAs) for trade operators in Mexico. Each product is interconnected within a unified platform to connect carriers and shippers and improve matching and operational efficiency via technologies such as live pricing, real-time tracking, digitization of critical documentation, brokerage support, transportation management, fleet management, and committed capacity solutions. For more information, please visit fr8technologies.com.

Forward-Looking Statements

This press release includes "forward-looking statements" within the meaning of the "safe harbor" provisions of the United States Private Securities Litigation Reform Act of 1995. Fr8Tech's and its subsidiary Fr8App Inc.'s actual results may differ from their expectations, estimates, and projections and, consequently, readers should not rely on these forward-looking statements as predictions of future events. Words such as "expect," "estimate," "project," "budget," "forecast," "anticipate," "intend," "plan," "may," "will," "could," "should," "believes," "predicts," "potential," "continue" and similar expressions (or the negative versions of such words or expressions) are intended to identify such forward-looking statements. Forward-looking statements in this press release include statements regarding the capabilities the Company is developing for its products and the anticipated application of its AI Lab research, including potential future applications of that research.

These forward-looking statements involve significant risks and uncertainties that could cause the actual results to differ materially from those discussed in the forward-looking statements. Most of these factors are outside Fr8Tech's and Fr8App Inc.'s control and are difficult to predict. Factors that may cause such differences include, but are not limited to: (1) the inability to obtain or maintain the listing of Fr8Tech's ordinary shares on Nasdaq; (2) changes in applicable laws or regulations, including laws and regulations governing artificial intelligence and data privacy; (3) the possibility that Fr8Tech or Fr8App Inc. may be adversely affected by other economic, business and/or competitive factors; (4) risks relating to the uncertainty of the projected financial information with respect to Fr8App Inc.; (5) risks related to the organic and inorganic growth of Fr8App Inc.'s business and the timing of expected business milestones; (6) risks related to the Company's development and use of artificial intelligence, including inaccurate or unexpected outputs, the possibility that research results obtained in controlled or simulated settings may not be replicated in commercial use, market acceptance of AI-enabled products, and reliance on third-party AI models and providers, whose availability, pricing or terms may change; and (7) other risks and uncertainties identified, including those under "Risk Factors," to be filed in Fr8Tech's filings with the Securities and Exchange Commission.

Fr8Tech cautions that the foregoing list of factors is not exclusive. Should one or more of these risks or uncertainties materialize, or should underlying assumptions prove incorrect, actual results may vary materially from those indicated or anticipated by such forward-looking statements. Fr8Tech and Fr8App Inc. caution readers not to place undue reliance upon any forward-looking statements, which speak only as of the date made. Fr8Tech and Fr8App Inc. do not undertake or accept any obligation or undertaking to release publicly any updates or revisions to any forward-looking statements to reflect any change in their expectations or any change in events, conditions or circumstances on which any such statement is based.



Fr8Tech Contact:

Jason Finkelstein
IGNITION Investor Relations
investors@fr8technologies.com

FAQ

AI-generated questions and answers. How Rhea-AI works. Not financial advice.

What did the Fr8Tech AI Lab study find about freight-rate negotiation agents?

The two best purpose-built negotiation engines won 32 of 36 statistically corrected comparisons and never lost against the nine language models tested. One reinforcement-learning engine closed deals on 62% of trading days in a balanced simulation but only 16% when competing for scarce demand.

How did Fr8Tech’s Spanish-language voice agents perform in simulated freight negotiations?

The direct speech-to-speech system misinterpreted less than 3% of spoken monetary amounts, versus approximately 13% for the two systems that relied on transcription. Across 231 recorded calls, it was also the only system that did not record an agreement the scripted carrier had not accepted.

Where can researchers get Fr8Tech’s negotiation-agent benchmarking toolkit?

Fr8Tech’s toolkit is available on GitHub under an Apache 2.0 license. Researchers and logistics companies can use it to test their own negotiation agents under the same protocol as the AI Lab.

Keep reading