Semantic search, vector retrieval, recommendation, semantic matching, knowledge bases, and retrieval-augmented generation
Type
Embedding
Context
384
Speed
7/10
Status
Current and accessible through Baidu Qianfan and AI Studio embedding APIs
Input
Â¥0.0005 per 1,000 input tokens
Output
Â¥0.0005 per 1,000 completion tokens in current model-list metadata; the model practically returns vector embeddings rather than generated completion text
View model
→
Multimodal understanding, long-context Chinese and English applications, complex reasoning, coding, tool-enabled agents, and enterprise workloads
Type
Multimodal
Context
249K
Reasoning
8/10
Speed
7/10
Multimodal
Image input
Audio input
Status
Current and accessible through Baidu Qianfan as of September 25, 2026
Input
¥0.006 per 1K input tokens for requests up to 32K tokens; ¥0.010 per 1K input tokens above 32K
Output
¥0.024 per 1K output tokens for requests up to 32K tokens; ¥0.040 per 1K output tokens above 32K
View model
→
Realistic text-to-image generation, reference-grounded visual creation, commercial-style imagery, and applications needing reduced generative-artificiality.
Type
Image Generation
Reasoning
2/10
Speed
8/10
Media output
Status
Available; listed in Baidu AI Cloud's inference-service model updates and supported for offline batch inference in Qianfan ModelBuilder documentation.
View model
→
Object removal, masked image repainting, image variation, and batch image-editing workflows.
Type
Image Editing
Reasoning
1/10
Speed
6/10
Multimodal
Image input
Media output
Status
Available with access dependent on Baidu Qianfan service configuration; listed for Qianfan batch inference. Related AI作画-iRAG版 API sales stopped on 2026-04-30, but that product notice does not explicitly state that the Qianfan ERNIE-iRAG-Edit endpoi
View model
→
Lightweight local text generation, Chinese and English language experimentation, compact conversational systems, and domain adaptation
Type
Lightweight
Context
131K
Reasoning
3/10
Speed
8/10
Fine-tuning
Streaming
Status
Current open-weight model
View model
→
Fast, low-cost Chinese text generation, long-context applications, enterprise agents, content creation, reasoning, and code assistance.
Type
General Purpose
Context
138K
Reasoning
7/10
Speed
9/10
Tool use
Web search
Status
Current and accessible through Baidu Qianfan; canonical API endpoint is ernie-4.5-turbo-128k.
Input
¥0.0008 per 1,000 input tokens; web-search augmentation is priced at ¥0.004 per 1,000 tokens where applicable.
Output
Â¥0.0032 per 1,000 output tokens.
View model
→
Cost-efficient multimodal understanding, image and video analysis, OCR, document comprehension, translation, visual question answering, and code-related tasks
Type
Multimodal
Context
128K
Reasoning
7/10
Speed
8/10
Multimodal
Image input
Video input
Status
Current and available
Input
Â¥0.003 per 1,000 tokens
Output
Â¥0.009 per 1,000 tokens
View model
→
Local or self-hosted multimodal applications, visual question answering, document and chart understanding, image analysis, video understanding, and efficient vision-language inference
Type
Multimodal
Context
131K
Reasoning
7/10
Speed
8/10
Multimodal
Image input
Video input
Status
Current open-weight model; publicly released under the Apache 2.0 license
View model
→
Agentic workflows, web-search-assisted tasks, reasoning, Chinese-language knowledge work, creative writing, and general-purpose assistant applications.
Type
General Purpose
Reasoning
8/10
Speed
8/10
Tool use
Web search
Status
Current; officially released and accessible through Baidu's ERNIE website and AI Studio playground. Public Qianfan API availability for an exact ERNIE 5.1 endpoint was not verified.
View model
→
Code completion, code generation, unit-test generation, code optimization, code explanation, and fine-tuning for software-development workflows.
Type
Coding
Context
128K
Reasoning
4/10
Speed
6/10
Fine-tuning
Status
Retired; Qianfan ModelBuilder listed August 14, 2025 as the retirement date.
View model
→
Chinese-language reasoning, long-form analysis, complex calculations, literary and document writing, agent workflows, and function calling
Type
Reasoning
Context
33K
Reasoning
8/10
Speed
8/10
Tool use
Web search
Streaming
Status
Ready; currently accessible through Baidu Qianfan as ernie-x1-turbo-32k
View model
→
Deep reasoning, Chinese and English question answering, mathematics, coding, factual responses, tool calling, web-grounded applications, and agent workflows
Type
Reasoning
Context
66K
Reasoning
8/10
Speed
7/10
Tool use
Web search
Streaming
Status
Current preview model
Input
Â¥0.001 per 1K input tokens
Output
Â¥0.004 per 1K output tokens
View model
→
Low-cost image-to-video generation, short social clips, product animation, marketing assets, and turning still images into dynamic scenes
Type
Video Generation
Reasoning
1/10
Speed
8/10
Multimodal
Image input
Media output
Status
Current and available through Baidu Qianfan API
Input
CNY 1.00 per 5-second video
View model
→
Chinese image-to-video generation, audiovisual storytelling, marketing videos, multi-person dialogue, synchronized speech, sound effects, and cinematic short-form content
Type
Multimodal
Speed
7/10
Multimodal
Image input
Media output
Status
Current model family; concrete Qianfan variants include Turbo, Lite, Pro, Turbo-I2V-Audio, and Turbo-I2V-Effect
View model
→
Low-cost text-to-image generation, marketing visuals, creative assets, illustrations, and rapid image prototyping
Type
Image Generation
Reasoning
1/10
Speed
8/10
Media output
Status
Current and available
Input
Â¥0.05 per image at 1024x1024
Output
Â¥0.05 per generated image at 1024x1024
View model
→
Multilingual OCR and structured parsing of complex documents, including tables, formulas, charts, seals, scanned pages, warped documents, and screen photographs
Type
Multimodal
Context
131K
Reasoning
3/10
Speed
8/10
Multimodal
Image input
Status
Available; superseded by PaddleOCR-VL-1.6
View model
→
Document parsing, OCR, layout analysis, table extraction, and structured document understanding
Multimodal
Image input
View model
→
Fast enterprise question answering, summarization, workflow nodes, agent response generation, and text processing with a 32K context window
Type
Lightweight
Context
33K
Reasoning
5/10
Speed
8/10
Tool use
Streaming
Status
Available through Baidu’s documented Wenxin Workshop API; not found in the current Qianfan V2 model-list documentation, so new integrations should verify legacy endpoint compatibility.
View model
→
Long-context Agent planning, task decomposition, component selection, and function-calling workflows.
Type
Lightweight
Context
128K
Reasoning
6/10
Speed
8/10
Tool use
Status
Current platform-listed planning model; standalone lifecycle status is not separately published.
View model
→
Enterprise agent workflows, intent recognition, instruction routing, and tool-calling tasks
Type
Other
Context
33K
Reasoning
4/10
Speed
8/10
Tool use
Status
Legacy; current public availability not verified
View model
→
Fast enterprise question answering, lightweight agent planning, component selection, and streaming text applications
Type
Lightweight
Context
8K
Reasoning
4/10
Speed
8/10
Tool use
Streaming
Status
Current and listed as available in Baidu Qianfan documentation; fast agent-oriented model
View model
→
