DeepSeek New Channels — Different models, different strengths

Different models · different strengths. One API, three discounted channels.

Flash for everyday work, Pro for harder jobs, Vision for screenshots. Same account, billed by channel discount.

Different models · Different strengths

Three DeepSeek lines

Pick the strength first, then the channel. Model IDs stay the same; discounts follow the channel.

DeepSeek V4 Flash

速度优先

适合速度和成本优先的代码任务。日常补全、小改动先用它。

日常编码和短任务先走这条。

FastEverydayCost

DeepSeek V4 Pro

复杂推理

适合复杂代码和深入分析。长链路任务再切这条。

复杂代码和深入分析再切这条。

CodingReasoningAgent

Flash 视觉实验版。看图改代码、读界面、处理带图工单。

截图、设计稿和带图的工单交给它。

VisionFastMultimodal

Model notes come from the current catalog. Live prices on this site take precedence.

Channel discounts

DeepSeek channel discounts

Same models, billed by channel. Check the discount, then pick the route.

01

战略算力资源

5.2 折

Routed via AWS/Azure cloud services, high stability for production environments

52% of list price

Choose a channel
02

第三方厂商

7.5 折

Routed via AWS/Azure cloud services, high stability for production environments

75% of list price

Choose a channel
03

官方直连

List price

Direct connection to official API, lowest latency and most reliable

Official list price, no extra discount

Choose a channel

Discounts come from the same source as channel management. Live prices on this site take precedence.

Public benchmarks

Capability and speed

Intelligence Index for capability, output speed for how fast it writes. Numbers come from the Artificial Analysis public API.

ModelIntelligenceCodingOutput speedFirst token
39.5
n/a
219 t/s
0.82s
36.0
68.8
80 t/s
1.12s
34.8
65.0
218 t/s
0.77s

Source: Artificial Analysis. Speed is median output tokens/s. Latency is time to first token in seconds. Source: Artificial Analysis

From one chat to your own workflow

Bring it into your workflow now.

Once the model is chosen, take the same Model ID into your tools in three steps.

1Get access

Create an API Key

Open your AI Code With account and create the key you'll use in your tool or application.

2Connect

Set Endpoint & Model

Use the OpenAI-compatible endpoint and paste in the exact Model ID.

https://api.aicodewith.ai/v1
3Make it real

Send your first request

Start with Flash, then switch to Pro or Vision when the task needs it.

Quickstart
Need the full walkthrough?
Follow the docs for copy-ready configuration and complete setup details.
Open Docs