Baidu API

CoBuddy is a code generation model from Baidu, released through the Qianfan platform and optimized for coding tasks and AI agent workflows. The model offers native support for both tool calling and reasoning, making it a strong fit for agentic use cases where the model needs to plan, invoke tools, and iterate. It provides a 131K token context window with up to 65K output tokens, giving it ample room for large codebases and extended generation. CoBuddy is engineered for high inference throughput and low end-to-end latency. It's a solid choice for developers building code-centric agents or assistive coding tools who need responsive performance alongside structured tool use.

Qianfan OCR Fast is a document intelligence model from Baidu's Qianfan team, purpose-built for optical character recognition tasks. It is an upgraded variant of the base Qianfan-OCR, trained on specialized OCR data while retaining general multimodal capabilities. The underlying Qianfan-OCR architecture is a 4B-parameter end-to-end vision-language model that replaces traditional multi-stage OCR pipelines with a single model handling document parsing, layout analysis, table extraction, chart understanding, key information extraction, and document QA. It performs direct image-to-Markdown conversion and supports 192 languages. The base model scored 93.12 on OmniDocBench v1.5 and 79.8 on OlmOCR Bench, leading all end-to-end models on both. Qianfan OCR Fast offers a 65K-token context window and is well suited for developers building document processing pipelines — invoice parsing, report extraction, exam grading, or RAG over scanned documents.

ERNIE 4.5 21B A3B is a lightweight text-only language model from Baidu using a Mixture-of-Experts architecture with 21B total parameters but only 3B active per token. It excels at general language understanding, generation, reasoning, and coding tasks while remaining computationally efficient. Released under Apache 2.0, it achieves competitive performance against larger models like Qwen3-30B-A3B despite having 30% fewer total parameters.

ERNIE 4.5 21B A3B Thinking is Baidu's reasoning-enhanced language model built on the 21B A3B architecture with explicit chain-of-thought capabilities. It activates only 3B of its 21B parameters per token while specializing in logic, mathematics, coding, and multi-step reasoning tasks. The model supports extended context up to 131K tokens and is optimized for complex problem-solving through structured thinking.

ERNIE 4.5 300B A47B is Baidu's flagship text-only large language model featuring 300B total parameters with 47B active per token via MoE architecture. It demonstrates state-of-the-art performance on instruction following and knowledge benchmarks like IFEval, SimpleQA, and ChineseSimpleQA. The model supports 131K context length and excels at text understanding, generation, reasoning, and coding.

ERNIE 4.5 VL 28B A3B is a lightweight multimodal vision-language model with 28B total parameters but only 3B active per token. It processes both images and text simultaneously, enabling tasks like image comprehension, chart analysis, document understanding, and cross-modal reasoning. The model offers both thinking and non-thinking modes while matching performance of larger models like Qwen2.5-VL-32B.

ERNIE 4.5 VL 424B A47B is Baidu's largest multimodal vision-language model with 424B total parameters and 47B active per token. It supports up to 131K context tokens and excels at visual reasoning, document/chart understanding, and visual question answering with both thinking and non-thinking modes. In thinking mode, it approaches or surpasses OpenAI o1 on reasoning benchmarks like MathVista, MMMU, and VisualPuzzle.