Realtime AI News
Cloudflare launches Clef, an open-source multimodal decision model built on Qwen
On October 1 local time, Cloudflare released Clef, an open-source multimodal decision model family built on Qwen, comprising the Clef and Clef-flash models and accompanied by a new reinforcement learning product for custom fine-tuning. Clef is based on Qwen3.8-27B and Clef-flash on Qwen3.5-9B, with a native vision encoder and a 64k context window.
Cloudflare has released Clef, an open-source multimodal decision model family built on Qwen, according to a report from IT之家 carried by ifeng on October 2. The launch, which took place on October 1 local time, includes two models, Clef and Clef-flash. The “decision” framing refers to the ability to judge and classify input states rather than to generate conversational text.
Cloudflare says Clef leads in the Jev Decision Index benchmark, evaluated under Decision Index 0.2.1. Both models are fully compatible with Jev-API, which lets developers test and verify them directly and gives existing applications built on Jev a comparatively smooth migration path.
On base models, Clef is built on Qwen3.8-27B, while the lighter Clef-flash uses Qwen3.5-9B. That pairing gives the family both a precision-oriented option and a version aimed at cost- and latency-sensitive workloads.
The main difference from the existing Jev is multimodality. Clef ships with a native vision encoder that can parse image inputs directly and classify visual content, whereas the current Jev supports text-only classification. That means Clef can take tasks with images built in rather than relying on a separate image-understanding step.
The context window has grown as well. Clef extends context to 64k, compared with 32k for Jev, allowing developers to load richer system input states for more comprehensive judgments.
Alongside the models, Cloudflare released a new reinforcement learning product. With it, customers can fine-tune Clef for their own business scenarios, aligning a general decision capability with their own rules.
Cloudflare says Clef offers high accuracy and strong overall performance, holding its own against or surpassing comparable models across several benchmarks. The specific comparisons rest on the benchmark table the company published, and real-world behaviour will still need to be validated against production workloads.
The choice to build on Qwen and ship open source continues a trend of pairing decision models with open ecosystems. For enterprises, the value of such models lies less in raw benchmark scores than in whether they can be embedded cheaply into existing content moderation, risk control or classification pipelines. What to watch next: the open-source licence, real deployment costs, and how useful the reinforcement learning fine-tuning tool proves in customer settings.
Why it matters
Clef brings multimodal classification and a 64k context to the open decision-model space, offering a lower-cost option for content moderation, risk control and classification workloads. By building on Qwen and bundling reinforcement learning fine-tuning, Cloudflare may further lower the engineering barrier for enterprises that want to train their own decision models.
Nearby Updates
All10/02, 13:20
OpenAI and Synopsys to Build GPT-Synopsys for Chip Design
Synopsys and OpenAI will jointly develop GPT-Synopsys, a specialised model optimised to operate Synopsys EDA tools and run semiconductor design workflows. The deal adds revenue sharing and a global go-to-market partnership, aiming to turn frontier models into native expert users of chip-design and verification tools.
10/02, 12:01
ServiceNow Introduces AutoSynthData for Enterprise Agent Training Data
ServiceNow AI has published AutoSynthData on the Hugging Face blog, presenting an approach for generating training data for enterprise agents. The work targets a familiar bottleneck: high-quality data for agentic systems is scarce and expensive to assemble by hand, making automated synthesis an increasingly important piece of the enterprise AI stack.
10/02, 14:43
Yokogawa joins Singapore AI data-centre testbed consortium
Yokogawa has joined a Singapore-based AI data-centre testbed consortium, according to IT Brief Asia. The move highlights how AI compute build-out is converging with data-centre infrastructure, and reflects Singapore's continued push to anchor the region's AI ecosystem.
10/02, 14:46
arXiv Caps Submissions at Two Per Month, With Rejections Still Counting
Starting October 1, 2026, arXiv limits each submitter to two paper submissions per calendar month across all categories, and rejected papers still consume the monthly quota. QbitAI reports that the change is meant to ease the review strain caused by a surge in submissions and low-value papers in the AI era.