deepseek-v3 · DeepSeek · text
The DeepSeek V3 series is a MoE large model with 671B parameters (37B active) and 128K context. It supports disk-based context caching, where a cache hit costs only about 7% of the input price, giving it a major cost advantage in long-context scenarios. Compatible with the OpenAI Chat Completions API.
128K context; disk caching (cache_hit ~26% of cache_miss); visual understanding; Function Calling; JSON mode; FIM completion; OpenAI API compatible; MoE architecture.
Cost-effective reasoning tasks; code generation and completion; long-document analysis; Chinese-language understanding; agent workflows; batch data processing; academic research and education.
Does not generate images; caching depends on hits (not guaranteed); high concurrency requires contacting sales; the FP8 quantized version is cheaper but with a slight accuracy loss.
Anthropic official CLI coding agent. Multi-file edits, tool use, long-context reasoning.
export ANTHROPIC_AUTH_TOKEN="YOUR_API_KEY"
export ANTHROPIC_BASE_URL="https://api.robovai.com"
export ANTHROPIC_MODEL="deepseek-v3"
claude