Connect with us

Technology

AutoTrust AI Releases JEV-27B, an Open Decision Model for Self-Hosted AI Agents

Published

on

Trained in about 9.2 hours on one NVIDIA B200, JEV-27B adds fast, calibrated System 1 decisions to a frozen Qwen3.8-27B backbone while preserving its System 2 generation path.

SINGAPORE, Sept. 29, 2026 /PRNewswire/ — AutoTrust AI today released JEV-27B, an Apache-2.0 open-weights model designed to handle frequent, structured decisions inside AI agent workflows while retaining the underlying model’s full generation and reasoning path.

JEV-27B answers yes/no, multiple-choice and 0–5 rating questions in a single forward pass and returns a calibrated probability for every option. It runs on one NVIDIA B200 inside a customer’s own infrastructure and serves both fast System 1 decisions and deliberate System 2 generation from a single set of weights.

The model trains a 108.9-million-parameter decision block—about 0.4% of the full model—on top of a frozen Qwen3.8-27B backbone. AutoTrust reports that training took approximately 9.2 B200-hours. With the decision block switched off, all 164 HumanEval completions were byte-identical to those produced by the base model.

“Jev proved there is real demand for models that decide rather than write,” said Daniel Tang, AutoTrust AI’s chief executive and co-founder. “JEV-27B shows that this capability can run on one GPU inside a customer’s own infrastructure, next to a reasoning model. For companies that cannot send every decision to a third-party API, that changes both the cost and the risk.”

Evaluation and evidence

AutoTrust AI evaluated JEV-27B across six public text-decision benchmark groups. It reported scores of 88.70% on JevBench, 83.75% on Kev, 73.89% on OpenJev text, 92.91% on Nimble, 77.46% on VitaminC and 87.71% on MASSIVE-en, for an equal-weight six-group mean of 84.07%.

For additional context, AutoTrust AI also ran the hosted TypeSafe Jev 1.13 API on the same benchmark groups and reported a six-group mean of 83.85%, with JEV-27B scoring higher on four groups and lower on two. Because AutoTrust conducted this comparison itself, the figures should be read as internal comparative evidence—not as an independent third-party validation or a claim of across-the-board superiority.

For public baseline context, AutoTrust reproduced the scores published by TokenRhythm for NeoHorse-Jev, Open-Jev, Kev and Laya English. AutoTrust did not rerun those four external baselines. The pinned source table is available at https://huggingface.co/TokenRhythm/NeoHorse-Jev-4B/blob/b50e043e22e0e41e7fc0c244e4daa707b8124930/README.md. These figures are included as published benchmark context rather than as a new same-environment comparison by AutoTrust.

JEV-27B was also evaluated for fidelity to its distillation target. On 25,376 held-out questions labeled with Jev 1.13 probability distributions, JEV-27B reported a mean KL divergence of 0.017, where zero means identical distributions. On decision-models-under-pressure, an independent benchmark scored against human labels, JEV-27B reached 96% of Jev 1.13’s accuracy with 16 answer options. In that independent test, JEV-27B approached—but did not exceed—Jev 1.13.

AutoTrust AI measured a median decision latency of 137 milliseconds and sustained throughput of about 130 decisions per second on one B200. The model card also cites third-party measurements of 238 to 301 milliseconds and 23 decisions per second for Jev’s hosted API. These are not controlled, like-for-like results: the hosted API measurements include network time, while AutoTrust AI’s local measurements do not, and the hardware, serving and concurrency conditions differ.

How it works

AutoTrust AI uses the terms System 1 for fast, typed decisions and System 2 for deliberate generation and reasoning. JEV-27B serves both from one set of weights. It follows JEV-9B as the company’s second integrated System 1 and System 2 open model.

The model is built with AutoTrust AI’s Blocks of Experts recipe. A strong pretrained model, Alibaba’s open-weights Qwen3.8-27B, stays frozen as one expert block. A small, detachable block is trained for a single skill, and a router sends each request either to the fast decision block or to the deliberate generation block.

The decision block holds 108.9 million trained parameters, 0.4% of the model, and took about 9.2 hours to train on one NVIDIA B200. The reasoning path was left untouched. With the decision block switched off, JEV-27B scores 78.0% on the HumanEval coding test, and all 164 of its completions are byte-identical to the base model’s.

In a demonstration reel released with the model, a self-hosted JEV-27B served as the decision engine for 10 tasks. It played Doom, making 64 decisions in a target-practice scenario, and steered a simulated drone through a MuJoCo obstacle course. It ran a live Google Flights search from Zurich to London and verified 21 results, navigated Wikipedia to Gödel’s incompleteness theorems, flagged four regression risks in a sample change to authorization code and routed a billing-refund ticket to support.

“Every AI agent is really a long chain of small decisions—which button to press, which file to open, which queue a ticket belongs in,” said Josh Liu, AutoTrust AI’s chairman and co-founder. “Make each one fast, private and cheap, and you change the economics of the whole chain.”

AutoTrust AI said JEV-27B inherits Jev 1.13’s blind spots, including multi-hop reasoning, arithmetic, dates and adversarial inputs, and that its training data is English-centric. The company says the model is not meant for high-stakes decisions and recommends gating its answers on confidence.

Benchmark context and source notes

Scores are in percent. JEV-27B and the hosted TypeSafe Jev 1.13 API were measured by AutoTrust AI; this is not third-party validation. NeoHorse-Jev, Open-Jev, Kev and Laya English are published baselines reproduced from TokenRhythm and were not rerun by AutoTrust. TokenRhythm source (pinned revision): https://huggingface.co/TokenRhythm/NeoHorse-Jev-4B/blob/b50e043e22e0e41e7fc0c244e4daa707b8124930/README.md. Full AutoTrust methodology and source notes: https://huggingface.co/autotrust/JEV-27B.

Availability

JEV-27B is available under the Apache-2.0 license at huggingface.co/autotrust/JEV-27B. The release includes the weights, decision adapter, training and serving code, vLLM support and full evaluation reports. A demonstration reel is available at https://huggingface.co/spaces/autotrust/JEV-27B-Demo.

AutoTrust AI plans to build the JEV decision block into future models in its Guru family, which powers the ScienceGuru research platform. ScienceGuru is available for Windows and macOS at https://scienceguru.ai/. AutoTrust AI also offers customized sovereign deployments for enterprises.

JEV-27B was trained on SargeDev/jev-distill-corpus-v3, a public, Apache-2.0-licensed corpus of Jev 1.13’s outputs. It shares no weights or code with, and is not affiliated with or endorsed by, TypeSafe AI. Jev and TypeSafe are trademarks of their respective owners.

About AutoTrust AI

AutoTrust AI Pte. Ltd. is a Singapore-incorporated AI research company building the Guru family of foundation models and ScienceGuru, an AI research platform for scientists and research teams. Its Blocks of Experts architecture combines pretrained expert blocks with small trained adapters to build frontier-capable and sovereign models efficiently. Learn more at autotrust.ai.

View original content to download multimedia:https://www.prnewswire.com/apac/news-releases/autotrust-ai-releases-jev-27b-an-open-decision-model-for-self-hosted-ai-agents-302891725.html

SOURCE AutoTrust AI

Continue Reading

Technology

YeahPay targets the trust gap holding back autonomous AI shopping

Published

on

By

SINGAPORE, Nov. 17, 2026 /PRNewswire/ — Yeahka Limited (9923.HK), a Hong Kong-listed payment technology company, is expanding its AI-driven merchant services and exploring new payment capabilities to address emerging challenges in agent-driven commerce, as artificial intelligence moves beyond providing recommendations towards executing transactions on behalf of consumers.

As AI agents increasingly participate in shopping activities, merchants face two practical challenges: making their products and services accessible to AI agents, and enabling transactions to be completed securely when purchases are initiated by AI rather than consumers themselves.

Yeahka is addressing these challenges through its merchant technology subsidiary, Fushi Tech, and its overseas payment brand, YeahPay.

Fushi Tech has launched Fynix AI Shop, an AI-powered merchant service platform deployed in selected overseas merchant scenarios. The platform supports customer engagement, personalised recommendations, appointment booking and order processing, with transactions completed through YeahPay.

Meanwhile, YeahPay is exploring payment protocols, identity verification and delegated authorisation mechanisms for AI agents, and has begun proof-of-concept collaborations with international financial institutions.

Together, these initiatives reflect Yeahka’s efforts to connect AI-powered merchant services with payment infrastructure as new models of commerce emerge.

Helping Merchants Adapt to AI-Driven Shopping

One of the challenges facing merchants in agent-driven commerce is ensuring that AI systems can accurately discover, understand and interact with their products and services.

Traditional merchant websites and e-commerce platforms are primarily designed for human users. However, AI agents operate differently, relying on structured product information, accessible interfaces and up-to-date data to identify relevant products, compare options and initiate transactions.

Samuel Chen, head of Fushi Tech, believes merchants will increasingly need to develop two separate interaction systems: “One for people and one for AI agents.”

While consumer-facing interfaces focus on browsing experience and usability, AI-facing systems must enable agents to retrieve product information, pricing and inventory data accurately, and facilitate subsequent transactions.

Based on this assessment, Fushi Tech is developing a merchant service framework tailored to AI agents. The company plans to help merchants establish independently controlled product and transaction platforms, using AI-friendly data structures and standardised interfaces to connect with major AI agents.

The approach is intended to improve the likelihood of merchants’ products being recognised, indexed and recommended by AI systems, while supporting the conversion of consumer enquiries into orders.

Fushi Tech has already begun applying AI to merchant operations through Fynix AI Shop. The platform integrates AI-powered customer interactions with commercial functions, allowing consumers to receive recommendations, make appointments and place orders through conversational interfaces.

As Amit Wakharkar, Fynix’s Director of Pre-sales and Solutions, noted, merchants do not simply need another question-and-answer-based customer service system. The more important objective is to convert customer interactions into tangible business outcomes.

Exploring Secure Payment Capabilities for AI Agents

Beyond product discovery and order processing, agent-driven commerce also raises new questions about payment authorisation and transaction security.

When an AI agent initiates a purchase on behalf of a consumer, payment providers must establish whether the agent is authorised to act, verify its identity and ensure that transactions remain within the permissions granted by the user.

At the MAG Payments Summit in Singapore on October 21, David Tay, Global Vice President of YeahPay, outlined three stages in the development of AI-powered commerce.

In the first stage, AI provides product recommendations, while consumers complete payments themselves. In the second, product discovery, ordering and payment take place within the same conversation, although consumers must still confirm each transaction. In the third stage, AI agents can autonomously execute purchases within authorisation limits established by consumers in advance.

According to Tay, as shopping activities become increasingly integrated into conversational interfaces, payment should become a natural part of the transaction process rather than a separate step that interrupts the user experience.

The third stage, in particular, will require payment infrastructure capable of supporting verifiable agent identities and secure delegated authorisation frameworks.

Against this backdrop, YeahPay is exploring payment protocols, identity verification and delegated authorisation mechanisms designed for agent-initiated transactions. The company has also begun proof-of-concept collaborations with international financial institutions to examine how these capabilities could support future AI-driven commerce.

These initiatives form part of Yeahka’s broader efforts to extend its payment capabilities to emerging transaction scenarios, while addressing the security and authorisation requirements associated with AI agents acting on behalf of consumers.

AI Agents Reshape the Commercial Landscape

The development of AI-powered shopping services is creating new possibilities for how consumers discover products, interact with merchants and complete purchases.

In September, Meta launched Muse, its personal AI agent, which can perform tasks including product searches, price comparisons and travel bookings, as well as complete purchases once authorised by users.

During Meta’s third-quarter earnings call on October 21, Mark Zuckerberg said that more than 10 million people in the United States and Canada had set up Muse within six weeks of its launch, with approximately one-quarter using it daily. Muse has since expanded into Japan and Singapore, integrating with local e-commerce and travel service providers.

To support AI-initiated purchases, Meta has integrated Stripe’s Link payment service into Muse, enabling transactions through payment methods linked to users’ accounts or through one-time virtual cards. Meta and other technology companies are also exploring interaction protocols intended to facilitate communication between AI agents and merchant systems.

These developments suggest that AI agents could gradually become another channel through which consumers access products and services. For merchants, this may create new opportunities to reach customers directly, while introducing additional requirements for product data, system compatibility and transaction processing.

However, AI-initiated purchases still represent a relatively small proportion of overall transactions. Challenges involving merchant system integration, delegated authorisation and payment security will continue to influence the pace of adoption.

For Yeahka, the emerging opportunity lies in connecting merchant-facing AI capabilities with secure payment services. Through Fushi Tech and YeahPay, the company is exploring how merchants can make their products and services accessible to AI agents while supporting transactions from product discovery and order placement through to payment completion.

As AI agents take on a greater role in commercial activities, the ability to connect merchant systems, AI interfaces and payment infrastructure will become an increasingly important consideration for the broader payments industry.

View original content:https://www.prnewswire.com/news-releases/yeahpay-targets-the-trust-gap-holding-back-autonomous-ai-shopping-302943090.html

SOURCE Yeahka

Continue Reading

Technology

Celebrate a Merry Giftmas with StarHub’s Festive HubDeals

Published

on

By

Dream phones up for grabs, plus festive treats and savings across Mobile, Broadband and Entertainment 

SINGAPORE, Nov. 18, 2026 /PRNewswire/ — This festive season, StarHub’s got you with more ways to give, save and stay connected through Merry Giftmas. From mobile upgrades to better connectivity at home and more entertainment for the family, Merry Giftmas brings together HubDeals across Mobile, Broadband and Entertainment from now until 4 January 2027.

Eligible mobile plans include HBO Max access, while selected 5Gbps and 10Gbps broadband plans include two WiFi 7 routers. HomeHub+ brings together TV and streaming options, with a Bang & Olufsen Home Entertainment Set included with an eligible package.

New and existing customers can also take part in weekly draws for a phone of their choice or a device voucher. StarHub will notify winners by email and announce them on its social channels after each draw.

StarHub’s Merry Giftmas festive HubDeals are available on the StarHub website, StarHub App and StarHub Shops.

For more information, visit starhub.com/promotions. Terms and eligibility criteria apply.

View original content to download multimedia:https://www.prnewswire.com/apac/news-releases/celebrate-a-merry-giftmas-with-starhubs-festive-hubdeals-302943092.html

SOURCE StarHub Ltd

Continue Reading

Technology

TIER IV contributes to native Autoware support in CARLA 1.0

Published

on

By

Simplifying autonomous driving development through open-source collaboration

TOKYO, Nov. 17, 2026 /PRNewswire/ — TIER IV, the pioneering force behind Autoware, open-source software for autonomous driving, has contributed to native Autoware support in CARLA 1.0, a major release of the open-source simulator for autonomous driving development and testing. Building on CARLA’s existing native ROS 2 interfaces, the new support enables Autoware to exchange sensor data, control commands and vehicle status directly with the simulator, without a separate Python-based bridge. This simplifies integration and reduces the software layers involved in simulation.

Advancing native integration between CARLA and Autoware

CARLA is an open-source simulator that enables developers to test autonomous driving systems in realistic virtual environments, reproducing roads, traffic conditions, and sensor data.

Early CARLA–Autoware workflows relied on a separate Python-based bridge, requiring additional setup and introducing overhead in data exchange. CARLA’s native Autoware support removes this intermediate layer, allowing developers to connect Autoware through CARLA’s launch configuration.

This streamlined approach simplifies development workflows and is expected to improve responsiveness and processing efficiency. It also supports the evaluation of both conventional modular autonomous driving systems and emerging AI-based approaches in simulation.

Expanding open-source collaboration

The native integration originated from discussions within the Autoware Foundation (AWF) community on establishing direct interoperability between Autoware and CARLA.

TIER IV began developing native CARLA–Autoware integration internally, and the CARLA team ported this work into CARLA 1.0. Its engineers also made sustained involvement in the CARLA open-source community and contributions to CARLA’s migration from Unreal Engine 4 to Unreal Engine 5, helping advance the simulator for the wider open-source community. These efforts helped establish a collaborative development process with CARLA maintainers and other contributors, supporting the implementation of native Autoware connectivity.

TIER IV plans to continue contributing improvements from its internal development, including optimized LiDAR point cloud output and enhanced vehicle models, to the upstream CARLA project.

Through this activity, TIER IV aims to make autonomous driving development and validation more accessible worldwide, working alongside contributors from AWF and the CARLA community.

Making autonomous driving more accessible

By simplifying integration between CARLA and Autoware, native support lowers technical barriers for developers building and testing autonomous driving systems. Alongside initiatives such as Autoware 2.0, accessible simulation environments can help expand the global Autoware ecosystem and accelerate the adoption of open-source autonomous driving technologies.

“Open-source software creates the foundation for the democratization of autonomous driving, but developers also need practical environments to build and validate real-world systems,” said Shinpei Kato, founder and CEO of TIER IV. “By contributing to CARLA’s development and native integration with Autoware, we are helping connect two important open-source communities. We look forward to expanding this collaboration and making autonomous driving development more accessible to organizations around the world.”

Availability and resources

Native Autoware support is included in CARLA 1.0release. Technical documentation, installation instructions, and implementation details are available through the following resources:

CARLA native ROS 2 documentationCARLA download and installation guideCARLA GitHub repositoryAutoware–CARLA integration discussionTIER IV’s CARLA–Autoware integration repository

About TIER IV
TIER IV stands at the forefront of deep tech innovation, pioneering Autoware, open-source software for autonomous driving. With a comprehensive suite of platforms and services built around Autoware, TIER IV provides everything from software development and vehicle procurement to operational support. Through the Autoware ecosystem, TIER IV works with partners worldwide to shape the future of intelligent vehicles with open-source software, aiming to create mobility that is safer, more sustainable, and accessible to all.

Media contact
pr@tier4.jp

The word marks and logos of TIER IV are trademarks or registered trademarks of TIER IV, Inc. in Japan and other countries. All other company names, product names, service names, and logos referenced herein are trademarks or registered trademarks of their respective owners.
Autoware is a registered trademark of the Autoware Foundation.

View original content to download multimedia:https://www.prnewswire.com/news-releases/tier-iv-contributes-to-native-autoware-support-in-carla-1-0–302942979.html

SOURCE TIER IV, Inc.

Continue Reading

Trending