· WPbyAI Research · Model Lab · 4 min read
DeepSeek V4 Pro 0813 Is Live: What WPbyAI Will Test for WordPress
DeepSeek now identifies 0813 as the current V4 Pro API build. Here is what changed, what is only an upstream claim, and what must pass before WPbyAI can use it.
DeepSeek now identifies DeepSeek-V4-Pro-0813 as the current version behind the deepseek-v4-pro API model name. WPbyAI has completed a documentation-level evaluation only: the model is not integrated, available to customers, or approved for production.
Evidence for this note was checked on August 13, 2026. DeepSeek’s official Models & Pricing page lists the 0813 build, while the official API change log has not yet added a separate 0813 entry. That distinction matters: a live model version is verifiable, but release-day performance claims still require controlled testing.
What the official documentation confirms
The public API model name remains deepseek-v4-pro. The resolved model version is now DeepSeek-V4-Pro-0813. DeepSeek currently documents:
- a one-million-token context window and a maximum output of 384K tokens;
- thinking and non-thinking modes;
- JSON output, tool calls, Responses API, and Anthropic-compatible API support;
- fill-in-the-middle completion in non-thinking mode;
- current prices of $0.003625 per million cached input tokens, $0.435 per million uncached input tokens, and $0.87 per million output tokens.
Those prices are dynamic. The same official page says an overall price increase is planned, so WPbyAI will record the dated price used in each benchmark rather than treating today’s rate as permanent.
DeepSeek’s earlier V4 preview release described V4 Pro as a 1.6-trillion-parameter mixture-of-experts model with 49 billion active parameters, a one-million-token context, and agent-oriented coding improvements. These are upstream specifications and claims, not WPbyAI test results for WordPress.
Why 0813 is relevant to WordPress delivery
A large context window could help an agent inspect a substantial theme, custom plugin, project brief, Template DNA, acceptance rules, and recent change history together. Tool calling and structured output are also relevant to a controlled workflow that must produce patches, manifests, validation results, and approval records.
None of those features proves that the model can safely modify WordPress. Context size does not guarantee that it will preserve approved facts, stay inside a file allowlist, use the correct WordPress APIs, avoid invented content, or recover cleanly after a failed edit. A low token price can also be erased by retries, excessive reasoning, broken builds, or human repair.
The WPbyAI evaluation gate
Before this model can move beyond the current EVALUATED — documentation only status, it must run the same versioned WPbyAI test contract used for other model candidates:
- convert an approved brief and design direction into a WordPress implementation without inventing business facts;
- preserve Template DNA through several connected changes instead of redesigning the site on every turn;
- load only the required WordPress engineering skills and use the correct APIs for the repository type;
- obey the workspace and file allowlist, produce a real diff, and fail safely on an empty or out-of-scope change;
- pass deterministic build, content, link, canonical, robots, sitemap, schema, accessibility, and critical-flow checks;
- return to the approved parent version after a rejected candidate without silently overwriting earlier work;
- report latency, token use, API cost, retry count, human-review minutes, and hard failures for the same fixed brief.
The first useful comparison is not a public leaderboard score. It is DeepSeek V4 Pro 0813 versus the current WPbyAI baseline on the same WordPress project, with the same inputs, tools, iteration budget, approval gates, and delivery standard.
A versioning lesson for model-based site builders
The stable API name did not change when the underlying build became 0813. That is convenient for integration, but it can also change behavior without a code deployment on the customer side. WPbyAI therefore treats the provider, requested model name, resolved model version when available, evaluation date, Recipe version, input manifest, and result evidence as separate facts.
If a provider does not return a resolved build identifier, the execution record must preserve the dated API configuration and public version evidence. A previously accepted benchmark is not automatically valid after an upstream model update.
Current WPbyAI status
| Item | Status on 2026-08-13 |
|---|---|
| Official API version observed | DeepSeek-V4-Pro-0813 |
| Documentation fit review | Completed |
| WPbyAI fixed-brief test | Not yet run |
| WordPress Skills/Recipe test | Not yet run |
| Customer pilot availability | No |
| Production use | No |
WPbyAI will publish a separate Lab result only after a reproducible WordPress run produces evidence. Until then, 0813 is a promising model candidate, not a supported product feature. Read how WPbyAI evaluates an AI WordPress builder after the first draft or request the current private-pilot boundary.