全面支持 Qwen 3.8 Max、DeepSeek-V4、GLM-5.3、Kimi K3、MiniMax M3、豆包 Seedance 2.5、可灵 Kling v3、万相 Wan 3.0，提供 100% OpenAI 兼容端点与端到端密码学验签存证。
DeepSeek
deepseek-v4-pro
DeepSeek next-gen MoE reasoning flagship for complex architectural design, autonomous coding loops, and math proofs.
Cached:
DeepSeek
deepseek-v4-flash
High-throughput, ultra-low latency DeepSeek model optimized for rapid tool calls and high-frequency agent loops.
Cached:
DeepSeek
deepseek-v4.1-flash
Upgraded 2026 DeepSeek lightweight model with enhanced reasoning density, sub-100ms first-token latency, and ultra-high throughput.
Cached:
Qwen
qwen3.8-max
Alibaba 2.4-trillion-parameter MoE flagship model with native 1M context, state-of-the-art autonomous coding, and comprehensive multimodal reasoning. Supports deterministic 0902 snapshot alias.
Cached:
Tencent
hy4-preview
Tencent Hunyuan 4 next-generation MoE frontier model with dual-speed reasoning, deep Chinese semantic comprehension, and enterprise agent tool integration.
Cached:
AI
glm-5.3-flash
High-speed lightweight GLM 5.3 model designed for real-time dialogue, extraction, and high-frequency agent tool execution.
Cached:
Qwen
wan3.0-video
Alibaba Wan 3.0 next-generation cinematic video generation model with photorealistic lighting, physics, and camera controls.
AI
glm-5.3
Zhipu AI flagship model with upgraded reasoning, deep tool invocation capabilities, and Chinese enterprise alignment.
Cached:
Kimi / Moonshot AI
kimi-k3
Moonshot AI flagship with ultra-long 2M lossless context, multi-document cross-referencing, and deep research synthesis.
Cached:
Kimi / Moonshot AI
kimi-k2.7-code
Specialized coding intelligence model from Moonshot AI with deep understanding of complex frameworks and full-stack tasks.
Cached:
MiniMax
minimax-m3
MiniMax flagship language and multimodal reasoning engine with native 512K context and expressive text generation.
Cached:
ByteDance
doubao-seedance-2.0
ByteDance flagship video generation model offering photorealistic cinematic camera movement and motion coherence.
ByteDance
doubao-seedance-2-0-260128
Accelerated Seedance 2.0 video generation engine with high rendering speed and crisp visual consistency.
ByteDance
doubao-seedance-2-0-mini-260615
Cost-optimized lightweight video generation endpoint tailored for social media clips and preview generations.
ByteDance
doubao-seedance-2-0-fast-260128
Low-latency video rendering endpoint with consistent motion dynamics for real-time video workflows.
ByteDance
seedance2.5
ByteDance frontier Seedance 2.5 video generation model with temporal consistency and complex scene choreography.
ByteDance
seedance2.5-huoshan
Official Huoshan direct enterprise channel for Seedance 2.5 with guaranteed compute rendering queues and zero concurrency drops.
Kling AI
kling-v3
Kuaishou Kling v3 video generation model featuring complex physics simulation and fluid real-world mechanics.
Kling AI
kling-v3-omni
Omni-modal video model synthesizing synchronized sound effects and motion pictures in one unified pass.
Kling AI
kling-video-o1
Reasoning-guided video generator ensuring narrative consistency and multi-shot continuity.
MiniMax
minimax-h3
MiniMax H3 video synthesis engine specializing in realistic character performance and natural facial expressions.
ByteDance
doubao-seedream-5-0-260128
ByteDance latest Seedream 5.0 high-resolution image synthesis engine with exceptional prompt fidelity and texture detail.
Qwen
qwen-image-3.0-pro
Alibaba Qwen Image 3.0 Pro model specialized in complex visual layout composition and text rendering.
100% compatible with OpenAI API specs. Drop straight into Next.js, LangChain, Cursor IDE, and LlamaIndex.
import { createOpenAI } from '@ai-sdk/openai';
import { streamText } from 'ai';
// Drop-in replace baseURL & apiKey
const batchin = createOpenAI({
baseURL: 'https://api.batchin.tech/v1',
apiKey: process.env.BATCHIN_API_KEY,
});
export async function POST(req: Request) {
const { messages } = await req.json();
const result = streamText({
model: batchin('deepseek-v4-pro'),
messages,
});
return result.toDataStreamResponse();
}