Overview
DeepSeek-V4-Pro-0813 is the current Pro model version shown on DeepSeek's official API pricing page. It should be understood as a formal Pro API snapshot in the DeepSeek V4 family, not as a separate product line or a new major-version launch blog. The April 24, 2026 DeepSeek V4 Preview introduced DeepSeek-V4-Pro and DeepSeek-V4-Flash with 1M context, open weights, and dual thinking modes; the current model table now identifies the Pro API model version as DeepSeek-V4-Pro-0813.
For users of DeepSeek, the practical change is clearer than the naming: Pro is now the active high-capability API target alongside Flash, with updated token pricing, a 384K maximum output limit, OpenAI-format and Anthropic-format endpoints, tool calls, JSON output, and Responses API support. Teams choosing between Flash and Pro should treat Pro as the stronger model for difficult reasoning, agentic coding, and long-context knowledge work, while using Flash when cost or concurrency matters more.
What's New
A Current Pro API Snapshot
The official DeepSeek pricing table lists deepseek-v4-pro with model version DeepSeek-V4-Pro-0813. That gives API users a concrete production target instead of relying only on the April V4 Preview label. The model still belongs to the V4 family and keeps the same official endpoint pattern: https://api.deepseek.com for OpenAI-format calls and https://api.deepseek.com/anthropic for Anthropic-compatible access.
1M Context and 384K Maximum Output
DeepSeek lists both V4 Flash and V4 Pro with a 1M-token context length and a maximum output of 384K tokens. That keeps Pro relevant for long-form codebase analysis, research synthesis, contract review, large-document workflows, and agent loops that need room for accumulated state. The output ceiling is unusually high for workflows that generate reports, plans, or multi-file edits in a single call.
Lower Pro API Pricing Than the Preview Page
The April preview content described V4-Pro at much higher API rates. The current official pricing table lists Pro at $0.003625 per 1M cache-hit input tokens, $0.435 per 1M cache-miss input tokens, and $0.87 per 1M output tokens. DeepSeek also notes that it plans to raise API pricing in the near future, so teams should verify the pricing page before committing a long-running workload.
Production API Features
DeepSeek-V4-Pro-0813 supports JSON output, tool calls, the Responses API, Anthropic API compatibility, chat prefix completion, and FIM completion in non-thinking mode. Thinking mode supports both non-thinking and thinking modes, with thinking as the default. For agent builders, the key value is not only model quality; it is being able to keep existing OpenAI or Anthropic client patterns while swapping the base URL and model name.
Compatibility Notes
DeepSeek's April V4 release note said deepseek-chat and deepseek-reasoner would be retired after July 24, 2026, and were then routing to V4 Flash non-thinking and thinking modes. New integrations should call deepseek-v4-pro or deepseek-v4-flash directly instead of relying on legacy aliases.
Existing OpenAI-compatible clients can usually migrate by updating the base URL and model ID. Anthropic-compatible clients should use DeepSeek's Anthropic-format base URL and verify streaming, tool-call, and response-shape assumptions before production rollout.
Pricing & Plans
| API Model | Cache-Hit Input | Cache-Miss Input | Output | Context | Max Output | Concurrency |
|---|---|---|---|---|---|---|
| deepseek-v4-pro | $0.003625 / 1M tokens | $0.435 / 1M tokens | $0.87 / 1M tokens | 1M | 384K | 500 |
| deepseek-v4-flash | $0.0028 / 1M tokens | $0.14 / 1M tokens | $0.28 / 1M tokens | 1M | 384K | 2500 |
Pricing is pay-as-you-go through the DeepSeek API. Cache-hit pricing is only relevant when your prompts reuse cached context. For cold prompts, Pro costs about 3.1x Flash on cache-miss input and output, while Flash provides five times the listed concurrency limit. DeepSeek warns that API prices may rise soon, so current cost estimates should be treated as time-sensitive.
Best For
- Teams that need DeepSeek's strongest current V4 API model for reasoning, coding, and long-context analysis
- Agent builders who want OpenAI-compatible or Anthropic-compatible API access with tool calls and JSON output
- Research and legal workflows that benefit from 1M context and very large generated outputs
- Cost-sensitive teams comparing Pro only for hard tasks while routing simpler calls to Flash
- Developers replacing legacy DeepSeek aliases with explicit V4 model IDs
FAQ
Is DeepSeek-V4-Pro-0813 a brand-new major version?
No. It is best treated as the current Pro model snapshot within the DeepSeek V4 family. DeepSeek's official April 24, 2026 post launched DeepSeek V4 Preview, while the current pricing table lists the live Pro API model version as DeepSeek-V4-Pro-0813.
What is the model ID for the API?
Use deepseek-v4-pro for the Pro API model. DeepSeek's pricing page lists its model version as DeepSeek-V4-Pro-0813, but the API model column shows the callable model as deepseek-v4-pro.
How much does DeepSeek-V4-Pro-0813 cost?
The official pricing page lists $0.003625 per 1M cache-hit input tokens, $0.435 per 1M cache-miss input tokens, and $0.87 per 1M output tokens for Pro. DeepSeek says it may raise API prices soon, so verify the current page before production use.
Should I use V4 Pro or V4 Flash?
Use Pro for difficult reasoning, agentic coding, and long-context work where accuracy matters more than raw cost. Use Flash when throughput, lower spend, and a higher listed concurrency limit matter more.
Does it support Anthropic-compatible calls?
Yes. DeepSeek lists both OpenAI-format and Anthropic-format base URLs for V4 Pro and V4 Flash. Teams should still test tool-calling and streaming behavior before switching production workloads.
