DeepSeek V4 Pro Launches: ¥3/M Tokens, 1M Context, Agent-Ready API
DeepSeek V4 Pro officially launches with aggressive pricing at ¥3 per million input tokens and ¥0.025 for cache hits, a 1M token context window, tool calling support, and compatibility with OpenAI and Anthropic APIs, positioning it as cost-effective infrastructure for AI agents and coding agents while highlighting a shift from benchmark competition to ecosystem competition.
DeepSeek V4 Pro Official Release
DeepSeek API has officially released the new model deepseek-v4-pro accessible via the endpoint DeepSeek-V4-Pro-0813. This release marks the model's entry into a developer-callable stage and signals a strategic push toward AI Agent, Coding Agent, and enterprise AI infrastructure.
Pricing: ¥3/M Input, ¥0.025/M Cache Hit, ¥6/M Output
Input tokens: ¥3 per million
Cache hit: ¥0.025 per million
Output tokens: ¥6 per million
The cache-hit price is especially significant. For Coding Agents and enterprise Agents, a single request often carries large amounts of repeated context — system prompts, tool definitions, project code, conversation history, and enterprise knowledge. A sufficiently low cache price directly reduces the actual running cost of agents, making large-scale agent deployment economically viable.
1M Context + Tool Calls: Foundation for Agent Infrastructure
DeepSeek V4 Pro supports a 1M token context window and Tool Calls . Together, these capabilities enable a shift from traditional RAG pipelines (retrieve → compress context → model → retrieve again) to feeding complete context directly to the model. In Coding Agents, an entire code repository, multiple files, modification history, and tool results can reside in a single task context. The model moves beyond answering questions to participating in a full understand → plan → call tools → execute → verify loop, which is key to putting agents into production.
API Compatibility: Responses API and Anthropic API Support
V4 Pro simultaneously supports Responses API and Anthropic API . This compatibility lowers switching costs for developers because many existing AI coding, agent, and MCP tools are already built around the OpenAI or Anthropic API ecosystems. Developers can swap models without redesigning their entire systems. This reflects a broader industry shift: competition among models is no longer just about benchmarks, but about API design, agent capabilities, and developer ecosystem. The easier a model is to integrate, the faster it enters real production environments.
New Criteria for Production-Ready Models
Beyond parameters, benchmarks, and reasoning ability, a model that can actually run in production now needs:
How long can the context be?
Is tool calling stable?
Is the API open?
Can it run agents?
Can it handle coding tasks?
Is the cost acceptable?
From this perspective, DeepSeek V4 Pro's launch represents DeepSeek's formal entry into the Agent Infrastructure layer. The model is only the first layer; a true AI Agent requires Model + Context + Tools + Memory + Runtime + Evaluation .
Multi-Model Management and the AI Infra Opportunity
With enterprises simultaneously using GPT, Claude, Gemini, DeepSeek, Qwen, and dozens of other models, a new problem emerges: how to unify access, management, and cost control across so many models?
This challenge is where the next real opportunity for AI infrastructure lies.
Key Specifications Summary
Model: deepseek-v4-pro Endpoint: DeepSeek-V4-Pro-0813 Input: ¥3 / million tokens
Cache hit: ¥0.025 / million tokens
Output: ¥6 / million tokens
Context window: 1M tokens
Supported APIs: Responses API / Anthropic API
The release underscores a transition: models provide intelligence, while AI infrastructure makes that intelligence production-ready.
Signed-in readers can open the original source through BestHub's protected redirect.
This article has been distilled and summarized from source material, then republished for learning and reference. If you believe it infringes your rights, please contactand we will review it promptly.
How this landed with the community
Was this worth your time?
0 Comments
Thoughtful readers leave field notes, pushback, and hard-won operational detail here.
