Skip to content
← Models
Open weightsBaidu/10 trustVerified
ERNIE 4.5 Turbo logo

ERNIE 4.5 Turbo

Baidu's speed-and-price-optimized multimodal flagship — frontier-class at a fraction of Western pricing.

Visit official page →

About

ERNIE 4.5 Turbo is Baidu's cost-optimized multimodal flagship, released April 2025. It handles text, images, audio, and video natively with a 128K-token context, and Baidu positions it against GPT-4.1-class multimodal work at roughly 40% of DeepSeek V3's cost — the company claims it beats GPT-4.5 on benchmarks at 1% of GPT-4.5's price (provider-claimed, treat accordingly). The ERNIE 4.5 base-model weights were open-sourced under Apache 2.0 in June 2025; Turbo is the hosted, price-optimized serving tier.

On Baidu's Qianfan platform it lists at about $0.55 input / $2.20 output per million tokens — an order of magnitude under Western frontier pricing. The ERNIE Bot consumer chatbot is free for registered users, and the model is deeply integrated across Baidu Search and the Wenxiaoyan app.

Specs

Context window

128,000 tokens

Pricing

$0.55 input / $2.20 output per 1M tokens on Qianfan (provider-published)

Free quota

ERNIE Bot chatbot free for registered users

API access

Baidu Qianfan (Wenxin Workshop) API; ERNIE 4.5 base-model weights on Hugging Face under Apache 2.0 (Turbo is the hosted variant)

Weights

Open weights

Released

Not disclosed

Benchmarks

Provider-reported: beats GPT-4.5 across benchmarks at 1% of the price.

Independent: no third-party reproduction found — treat all figures as provider-claimed.

Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.

Use cases

multimodal-chatbotschinese-language-aidocument-analysiscost-efficient-inference

Trust score — 7/10

/10 trustVerified

Official Qianfan product page verified live (HTTP 200); Apache 2.0 base weights independently downloadable on Hugging Face. Docked for: (1) all capability-vs-GPT-4.5 claims are provider-run; (2) the Turbo tier itself is API-only, so the open-weights story applies to the base family, not this exact variant; (3) international onboarding is friction-heavy.

More in the ernie family

Sources