What is the Daily-Omni benchmark and what does it test?

Daily-Omni is a third-party benchmark designed to evaluate a model's ability to understand and reason across audio and visual information in everyday situations. It contains 684 real-world videos and 1,197 multiple-choice questions covering six task categories, including audio-visual alignment, event sequencing, inference and reasoning.

How does WITA-Omni differ from conventional robotic systems?

WITA-Omni extends the "Thinker-Talker" framework with an additional "Actor" component that treats movement and facial expressions as native outputs alongside speech. This unified architecture allows the Thinker (reasoning engine), Talker (speech generation), and Actor (physical movement and facial expressions) to operate within a shared model state, letting robots continue observing their surroundings while preparing and delivering responses.

How was WITA-Omni trained?

WITA-Omni was trained on tens of millions of hours of multimodal data during continued training, expanding its capabilities to include audio perception and cross-modal reasoning. The training process used three stages: supervised fine-tuning, on-policy distillation, and reinforcement learning (using Group Relative Policy Optimization to improve interaction decisions).

Large Language ModelsAudio & SpeechRobotics & Automation NewsAug 3, 2026Aug 3, 2026, 19:00 JST

AGIBOT's audio-visual model tops Daily-Omni benchmark at 85.21%

Send on LINE

3 Key Points

What happened
AGIBOT's WITA-Omni Preview foundation model achieved an average accuracy of 85.21 percent on the Daily-Omni audio-visual reasoning benchmark, outperforming competitors including Alibaba's Qwen3.5-Omni-Plus, Google's Gemini 3.1 Pro Preview, and ByteDance's Doubao Seed 2.0 Lite. The model ranked first or joint first in six of the eight reported evaluation metrics.
Why it matters
The Daily-Omni benchmark tests whether AI models can understand audio and visual information together in real-world situations—a capability AGIBOT says is essential for robots operating in dynamic environments, where they must associate sounds with actions, follow event sequences, and decide whether and when to respond. Unlike image-text benchmarks, Daily-Omni evaluates cross-modal reasoning across audio, video, and their combination.
What to watch
AGIBOT's WITA-Omni Preview integrates a "Thinker-Talker-Actor" architecture that generates speech and coordinated physical movements and facial expressions in real time. The company plans to integrate this technology into its robotic platforms, including humanoid robots and quadrupeds, to enable more natural human-robot interactions.

Not sure about something? Ask the AI

Questions and answers are published on this page.

Summaries like this, in your inbox every morning.

Get news like this every morning, free

Context & Analysis

AGIBOT's result on Daily-Omni reflects a shift in how foundation models for robotics are being evaluated. Traditional AI benchmarks emphasize image-text understanding, but Daily-Omni tests cross-modal reasoning—the ability to link sounds to visible events and understand how situations develop over time. This distinction matters because embodied AI systems (robots operating in physical environments) must interpret multiple data streams simultaneously. A robot that hears a voice but cannot connect it to the person speaking, or that misses the sequence of events unfolding around it, will struggle to interact naturally or make safe decisions.

AGIBOT's approach—training on tens of millions of hours of multimodal data and then refining the model with thousands of hours of human-centric interaction recordings—addresses this gap. The company's emphasis on temporal relationships (the timing and order of audio, video, movement, and facial expressions) suggests that learning WHEN and HOW to respond matters as much as learning WHETHER to respond. This is consistent with AGIBOT's stated design goal: not only to generate accurate responses but also to determine appropriateness, timing, and intended recipient.

FAQ

What is the Daily-Omni benchmark and what does it test?: Daily-Omni is a third-party benchmark designed to evaluate a model's ability to understand and reason across audio and visual information in everyday situations. It contains 684 real-world videos and 1,197 multiple-choice questions covering six task categories, including audio-visual alignment, event sequencing, inference and reasoning.
How does WITA-Omni differ from conventional robotic systems?: WITA-Omni extends the "Thinker-Talker" framework with an additional "Actor" component that treats movement and facial expressions as native outputs alongside speech. This unified architecture allows the Thinker (reasoning engine), Talker (speech generation), and Actor (physical movement and facial expressions) to operate within a shared model state, letting robots continue observing their surroundings while preparing and delivering responses.
How was WITA-Omni trained?: WITA-Omni was trained on tens of millions of hours of multimodal data during continued training, expanding its capabilities to include audio perception and cross-modal reasoning. The training process used three stages: supervised fine-tuning, on-policy distillation, and reinforcement learning (using Group Relative Policy Optimization to improve interaction decisions).

Robotics & Automation NewsRead Original Article

Send on LINE

AI news that matters for your work, delivered every morning.

Pick your industry and the AI tools you use, and get news related to your work every day.

Get it free

Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →

Ask AI

Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.