DeepSeek V3

im/together-deepseek-v3

stable

DeepSeek's 671B MoE flagship. Trained in China — different alignment lens than US labs.

Specs

Capabilities and limits

Provider: Together AI
Model family: deepseek-v3
Type: chat
Context length: 131,072 tokens
Max output: 8,192 tokens
Streaming: Yes
Tool calling: Yes
JSON mode: Yes
Vision: No

Pricing

Active credit rate

Input: 1500000 cr / M tokens · $1.5000
Output: 1500000 cr / M tokens · $1.5000
Effective from: 5/17/2026

Best for

reasoningcodingmathlong-context

Endpoints

Routes this model serves

/v1/chat/completions