VIDRAFT
PulseAugur coverage of VIDRAFT — every cluster mentioning VIDRAFT across labs, papers, and developer communities, ranked by signal.
- 2026-08-02 research_milestone VIDRAFT achieved the #1 verified result in the Google x Hugging Face Fast Gemma Challenge for optimizing the Gemma 4 E4B it model's speed while maintaining quality. source
- 2026-07-13 product_launch VIDRAFT's Darwin model family achieved over 1 million cumulative downloads on Hugging Face. source
11 day(s) with sentiment data
-
South Korea funds national cybersecurity AI model; SKT, NAVER consortia bid
South Korea's Ministry of Science and ICT is funding a program to develop a national cybersecurity-specialized AI foundation model. Two major consortia, one led by SKT and another by NAVER Cloud, have submitted proposal…
-
Korean AI Startup VIDRAFT Releases PROMETHEUS Embodied World Model
VIDRAFT, a Korean AI startup, has released PROMETHEUS, an embodied AI world model designed for simulating and reasoning about physical environments. This model aims to advance embodied intelligence and world modeling, c…
-
AI startup VIDRAFT unveils AX-RAY to detect hidden causal leakage in models
VIDRAFT, an AI safety startup, has developed a new auditing framework called AX-RAY to detect causal leakage in hybrid sequence models. This leakage occurs when future token information improperly influences earlier pos…
-
LLMs including GPT, Claude, Gemini face drug discovery benchmark
VIDRAFT has launched the Open Discovery Challenge, a new public benchmark designed to compare the drug candidate discovery capabilities of leading large language models including OpenAI's GPT series, Anthropic's Claude,…
-
Claude models lead early drug discovery challenge against GPT-4
VIDRAFT has launched the Open Discovery Challenge, a drug discovery competition hosted on Hugging Face, where participants use various AI models to propose candidate molecules for malaria and tuberculosis. Early data fr…
-
VIDRAFT launches Hansu, a browser-based local AI prioritizing data sovereignty
VIDRAFT, a Korean startup, has launched Hansu, a browser-based AI assistant designed for local, client-side inference. This approach prioritizes data sovereignty, ensuring user data remains within the browser and is not…
-
VIDRAFT releases 398B parameter LLM specialized for scientific reasoning
VIDRAFT, a Korean AI startup, has published evaluation results for its new 398-billion parameter model, Darwin-398B-JGOS. This large language model is specifically designed for scientific reasoning tasks, aiming for rig…
-
Korean AI models face scrutiny over 'borrowed' tech and narrow benchmarks · 1 source tracked
A viral debate on China's Zhihu platform scrutinized a wave of Korean AI models that claimed to outperform DeepSeek. The critique focused on several points: models allegedly building upon open-weight models from other l…
-
VIDRAFT team achieves verified SOTA in Fast Gemma Challenge
The VIDRAFT team, competing as vidraft-darwin, achieved a verified state-of-the-art result in The Fast Gemma Challenge by optimizing the Google Gemma model. Their submission, vidraft-fw188-ctk49-n64-patchbridge-v1, reac…
-
VIDRAFT tops Google x Hugging Face Fast Gemma Challenge with verified speed and quality
VIDRAFT, an AI agent developed by a Korean startup, has achieved the top verified position in the Google x Hugging Face Fast Gemma Challenge. The challenge focused on optimizing the performance of Google's Gemma 4 E4B i…
-
Korean startup's Darwin-398B-JGOS model ranks 3rd globally on GPQA Diamond
A South Korean startup, VIDRAFT, has developed a language model named Darwin-398B-JGOS that achieved the top rank among Korean models on the GPQA Diamond benchmark. This model, reportedly trained on approximately 24 GPU…
-
Korean startup VIDRAFT achieves AI breakthroughs with 24 GPUs
South Korean startup VIDRAFT has achieved significant results using a modest setup of approximately 24 GPUs, challenging the dominance of Big Tech's large-scale operations. The company's Darwin models have topped the Ko…
-
LLM inference engines see major speed boosts with new techniques and optimizations · 4 sources tracked
Several projects are enhancing the performance of local LLM inference engines. Kernel Acceleration (VK) has developed three engines (VKAE, VKUE, VKIE) that improve token generation speed by up to 601 tokens per second t…
-
Aether-7B-5Attn leads charge for truly open LLMs with full transparency
Aether-7B-5Attn, a 7B Mixture-of-Experts model from VIDRAFT, has been added to the awesome-free-models list, distinguishing itself from other "open" models. Unlike typical "open-weight" models which only provide trained…
-
VIDRAFT's POCKET 35B model runs on phones and PCs without GPUs
VIDRAFT has released POCKET, a 35 billion parameter Mixture-of-Experts model that can run on devices without GPUs, including PCs and phones, using standard tools like llama.cpp. This offers an alternative to models like…
-
Aether-6B-11Attn-base: Mid-training model released as research artifact
A new research artifact, Aether-6B-11Attn-base, has been released mid-training, offering a unique look into the development process of a model that combines eleven heterogeneous mixers including conventional attention, …
-
VIDRAFT releases fully open-source Aether-7B-5Attn model
VIDRAFT has released Aether-7B-5Attn, a fully open-source foundation model under the Apache 2.0 license. Unlike many "open" models that only provide weights, Aether-7B-5Attn includes its architecture, training data reci…
-
VIDRAFT's Darwin model family hits 1M Hugging Face downloads via evolutionary merging
VIDRAFT's Darwin model family has surpassed one million downloads on Hugging Face, driven by its unique training-free evolutionary merge methodology. This approach allows for rapid iteration and recombination of model s…
-
VIDRAFT ships dual LLM serving engines for GPU throughput and CPU reach
VIDRAFT has developed two distinct serving engines for large language models, addressing separate optimization targets. VKAE is a kernel-level acceleration engine designed to maximize throughput on GPUs, achieving up to…
-
Tencent and VIDRAFT showcase sparse MoE models with reduced active parameters
Tencent has released Hy3, a 295-billion-parameter Mixture-of-Experts (MoE) model that utilizes only 21 billion active parameters per forward pass, significantly reducing inference costs. This MoE architecture, featuring…