Baidu unveils Ernie 5.0, an AI model to process and generate text, images, audio, and video, claiming it beats GPT-5-High and Gemini 2.5 Pro on some benchmarks
our latest natively omni-modal foundational model. It excels in omni-modal understanding, creative writing, instruction following, and more. We will continue investing in and developing more cutting-edge models to push the boundaries of intelligence. [image] Lei Li / @_tobiaslee : ERNIE 5.0 is out! Truly Amazing with three highlights, and here are my comments: Natively Omni-Modal with Early Fusion Architecture: incorporation of text, images, video, and audio modalities from the initial training phase. -> This is really awesome for user experience - being [image] Rohan Paul / @rohanpaul_ai : 🚨 MASSIVE. Baidu just launched ERNIE 5.0, a 2.4T-parameter natively omni-modal MoE. Handles text, images, audio, and video in 1 unified model while only activating under 3% of its parameters on each step. - Across more than 40 authoritative benchmark evaluations, its language [image]
VentureBeatCarl Franzen
Context & Ripple Effects
Baidu has repeatedly advanced the Ernie line through model releases and comparative performance claims, from Ernie 4.0’s claimed GPT-4-level capabilities to a free reasoning model and upgraded flagship in 2025. Ernie 5.0 extends that cadence from language and reasoning toward a single model spanning text, images, audio, and video.
The reported early-fusion, mixture-of-experts design makes the release notable beyond benchmark positioning: Baidu is presenting multimodal generation and understanding as a core foundation-model capability rather than an add-on.
First-order effects
Baidu gains a new flagship with which to pitch multimodal AI workloads, while its claims against GPT-5-High and Gemini 2.5 Pro raise the bar for how Ernie will be evaluated.
Developers and enterprise buyers considering Ernie can assess one model for inputs and outputs across text, images, audio, and video rather than treating those modalities as separate model categories.
Second-order effects
Competing model providers face added pressure to substantiate multimodal benchmark claims and demonstrate whether their systems handle cross-media tasks coherently, not merely support multiple formats.
The sub-3% parameter activation claim puts attention on efficiency alongside model quality, reinforcing buyer scrutiny of Baidu’s earlier free Ernie X1 and Ernie 4.5 release and other providers’ cost-to-capability trade-offs.
Third-order effects
If native multimodal architectures prove useful in production, model competition will shift from standalone language performance toward integrated media workflows and the economics of serving them.
That shift could increase buyer leverage: enterprises may favor providers that combine broad modality coverage, credible task performance, and efficient inference over models marketed on a single benchmark lead.
The trend: Foundation-model competition is moving toward natively multimodal, efficiency-conscious systems designed to serve more end-to-end AI workflows.
The newly released ERNIE 5.0 is really interesting. I believe the “natively multimodal architecture” will fundamentally redesign how AI perceives the world. Models are evolving from “think with text” to “think with world”. ERNIE 5.0 is a compelling attempt at this. Congrats!🥳 [im…
@hongruwang007
Hongru Wang on x
ERNIE 5.0 is here, and it's redefining what AI can do! 📷 From seamless multimodal magic (text+image+audio = endless possibilities) to deep, nuanced understanding (it gets the joke and the subtext), plus human-like generation that feels like a real conversation. It is so happy to …
@rohanpaul_ai
Rohan Paul on x
🧵2/n. What makes ERNIE 5.0 stand out is that it brings a unified omni-modal approach that blends text, visuals, sound, and video into a single system for both understanding and creation. Here “natively omni-modal” basically means ERNIE 5.0 was built from the ground up to handle […
@socialwithaayan
Muhammad Ayan on x
OMG... Baidu just dropped ERNIE 5.0 a natively omni-modal large model and the results feel unreal. It matches top global models and reaches world-leading levels in multimodal generation. Nothing else right now feels this complete. Full breakdown in thread: 🧵 [image]
@kimmonismus
@kimmonismus on x
Ernie 5.0 has been released, and it's a bigger leap than you might think! A global battle for computing power continues. However, China remains undeterred and is relying on its own resources. While it's likely they received some chips via Singapore and probably a few of the [imag…
@poezhao0605
Poe Zhao on x
Baidu's World Conference today was all about applications. The new model launch? That came last. ERNIE 5.0 dropped with 2.4 trillion parameters using a sparse MoE architecture. Only 3% of parameters activate per inference, though Baidu hasn't disclosed the exact expert count.
@s_mohinii
Mohini Shewale on x
Finally! Baidu has officially introduced ERNIE 5.0. As a natively omni-modal model, it delivers real omni-modal understanding and generation across text, images, audio, and video. 🚀 Here's what makes it stand out: ✅ Powered by a 2.4T-scale MoE architecture, with under 3% [image]
@n1d_all
@n1d_all on x
@Baidu_Inc ERNIE 5 preview is better than GPT 5.1 in creative writing. I love it so much. [image]
@kimmonismus
@kimmonismus on x
ERNIE 5.0 stands out by combining a unified omni-modal model with a highly efficient 2.4T-parameter MoE architecture to deliver frontier-level performance across more than 40 benchmarks while keeping compute remarkably low. And this is despite all the challenges China faces. [vid…
@baidu_inc
@baidu_inc on x
Here comes ERNIE 5.0 — our latest natively omni-modal foundational model. It excels in omni-modal understanding, creative writing, instruction following, and more. We will continue investing in and developing more cutting-edge models to push the boundaries of intelligence. [image…
@_tobiaslee
Lei Li on x
ERNIE 5.0 is out! Truly Amazing with three highlights, and here are my comments: Natively Omni-Modal with Early Fusion Architecture: incorporation of text, images, video, and audio modalities from the initial training phase. -> This is really awesome for user experience - being […
@rohanpaul_ai
Rohan Paul on x
🚨 MASSIVE. Baidu just launched ERNIE 5.0, a 2.4T-parameter natively omni-modal MoE. Handles text, images, audio, and video in 1 unified model while only activating under 3% of its parameters on each step. - Across more than 40 authoritative benchmark evaluations, its language [im…