Large Language Models
236B Mixture-of-Experts model (21B active) combining general chat and DeepSeek-Coder-V2 capabilities with OpenAI-compatible API.
DeepSeek V2.5 unifies DeepSeek-V2-Chat and DeepSeek-Coder-V2-Instruct into a single 236B-parameter Mixture-of-Experts (MoE) model that activates only 21B parameters per token, delivering frontier-grade coding and math reasoning at a fraction of dense-model inference cost.
base_url to https://api.deepseek.com with model deepseek-chat—zero SDK rewrite required.| Nom de l'outil | DeepSeek V2.5 |
|---|---|
| Catégorie | Large Language Models |
| Modèle tarifaire | Open Weights / API ($0.14 / 1M in) |
| Tag thématique | DeepSeek, MoE LLM, Open Weights, Cheap LLM API |
Signaux de trafic, de classement et d’engagement pour cet outil.
Aucune donnée analytique pour le moment
DeepSeek's OpenAI-compatible API charges approximately $0.14 per 1M input tokens on cache miss, just $0.014 per 1M on automatic disk KV cache hit (90% off), and $0.28 per 1M output tokens—over 95% cheaper than frontier closed models.
DeepSeek V2.5 unifies DeepSeek-V2-Chat and DeepSeek-Coder-V2 into a single 236B-parameter Mixture-of-Experts (MoE) model with Multi-head Latent Attention (MLA), activating only 21B parameters per token across a 128K context window.
Yes. You only need to change `base_url` to `https://api.deepseek.com` and pass your DeepSeek API key in existing OpenAI Python/Node SDKs, Cursor, or LangChain apps.
Yes—model weights are publicly available on Hugging Face under the DeepSeek Model License (permitting commercial use), though its 236B total MoE size typically uses multi-GPU server nodes, making the ultra-cheap hosted API preferred for most teams.
Copy the HTML snippet below to display your ToolAI featured badge on your website or docs.
