ERNIE X1 Turbo

What it is

ERNIE X1 Turbo is a reasoning model from Baidu, launched on 25 April 2025 at Baidu Create 2025 in Wuhan^[1]. Baidu describes it as an upgraded "deep-thinking" reasoning model: compared with what came before, it uses a more advanced chain of thought and has stronger multimodal and tool-invocation abilities^[1]. Chain of thought here means the model works through a problem in intermediate steps before giving an answer, rather than producing a reply in one pass; that extra working-through is what reasoning models are built around. Multimodal ability means handling more than text, and tool invocation means calling outside functions or services while answering. The model is free to use on ERNIE Bot, Baidu's consumer-facing chatbot^[1].

Where it came from

X1 Turbo is an upgraded version of ERNIE X1, which first appeared on 16 March 2025^[1]. Baidu framed the Turbo generation — ERNIE 4.5 Turbo together with X1 Turbo — as a response to four problems it saw in existing models: being limited to a single modality, high hallucination rates, slow responses and high prices^[1]. Hallucination means the model stating things that are not true with apparent confidence; latency and price are the practical costs of running a large model.

What it is good at

Baidu's release says X1 Turbo excels at question answering, literary creation and logical reasoning^[1]. Coverage of the launch described it as suited to literary creation, complex logical reasoning, code generation and following intricate instructions^[2]. Those are the tasks where a model has to hold several constraints in mind at once, which is where an explicit step-by-step reasoning process is supposed to help.

Performance claims

Baidu claimed that X1 Turbo outperforms DeepSeek-R1 and the latest DeepSeek-V3^[1]. The company did not publish benchmark scores to support that claim^[1], so it should be read as a vendor statement rather than a measured result that outside parties can check.

Price

Pricing was a central part of the announcement. Baidu set the API price at RMB 1 per million input tokens and RMB 4 per million output tokens^[1]. That is half the price of ERNIE X1 and 25% of DeepSeek-R1's price^[1]. Tokens are the small chunks of text a model reads and writes, and they are the usual unit for billing API access: input tokens are the prompt sent to the model, output tokens are what it generates. Because a reasoning model writes out its intermediate steps, it typically produces more output tokens per request than a model that answers directly, which makes the per-token output price matter more than it otherwise would.

Where it sits in the line

X1 Turbo is the successor to ERNIE X1, and it was itself followed by ERNIE X1.1, unveiled on 9 September 2025^[1][3]. The sequence — X1 in March, X1 Turbo in April, X1.1 in September — shows Baidu iterating on its reasoning line within a single year, with the Turbo step focused on capability gains alongside a sharp cut in API price^[1].