AI Models Positive 7

DeepSeek V4 Pro 0813 GA: 1 endpoint for OpenAI Responses API drop-in

DeepSeek's GA release of V4 Pro 0813, a large-scale mixture-of-experts model, now supports the OpenAI Responses API for drop-in Codex use. The compatibility layer silently ignores unsupported parameters and uses SSE streaming without `data: [DONE]`. OpenRouter offers single-provider hosting with performance and uptime tracking.

· 4 min read · Verified by 2 sources ·

AI briefing

Key takeaways

7 impact
Positivesentiment
2sources
4min read
  1. DeepSeek's GA release of V4 Pro 0813, a large-scale mixture-of-experts model, now supports the OpenAI Responses API for drop-in Codex use.
  2. The compatibility layer silently ignores unsupported parameters and uses SSE streaming without `data: [DONE]`.
  3. OpenRouter offers single-provider hosting with performance and uptime tracking.
Drawn from
  • Hacker News
  • openrouter.ai

In this briefing

Mentioned

Key Intelligence

Key Facts

  1. 1DeepSeek V4 Pro 0813 is described by OpenRouter as a "large-scale mixture-of-experts model" and "the GA release of DeepSeek V4 Pro."
  2. 2DeepSeek's API now supports the OpenAI Responses API format with base_url https://api.deepseek.com, enabling DeepSeek models in Codex via the OpenAI SDK.
  3. 3Python integration requires `pip3 install openai` and setting `base_url = "https://api.deepseek.com"`; the example calls model `deepseek-v4-flash`.
  4. 4Streaming uses semantic server-sent events (SSE) with monotonically increasing `sequence_number` and ends with `response.completed`, `response.incomplete`, or `response.failed` instead of `data: [DONE]`.
  5. 5Unsupported Responses API parameters are silently ignored; response fields `store` is fixed to `false`, `previous_response_id` to `null`, and `parallel_tool_calls` to `true`.
  6. 6OpenRouter hosts DeepSeek V4 Pro 0813 through one provider, forwarding every request directly with no routing decisions, and tracks throughput, latency, TTFT, and 30-day uptime.

DeepSeek V4 Pro 0813

Company
Architecture
Mixture-of-experts
Release
GA
Model Slug
deepseek/deepseek-v4-pro-0813
Base Url
https://api.deepseek.com

Analysis

For ML engineers and AI product teams, GA releases are the moment experimental models become production candidates. DeepSeek V4 Pro 0813 arrives with a deliberate interoperability move: one OpenAI-compatible Responses API endpoint that lets existing Codex and OpenAI SDK stacks integrate without modification. The technical details that matter are now visible in the docs — unsupported parameters are silently ignored, store is fixed to false, previous_response_id is null, and parallel_tool_calls defaults to true.

DeepSeek has made a quiet but strategically significant move in the AI infrastructure market: the availability of DeepSeek V4 Pro 0813, described by OpenRouter as the GA release of DeepSeek V4 Pro, alongside new API documentation showing OpenAI Responses API compatibility. The docs, published on api-docs.deepseek.com, explicitly state that support was added "to meet the demand for Codex," and show how to use DeepSeek models with an unmodified OpenAI SDK by changing only base_url to https://api.deepseek.com and using the API key. This positions DeepSeek to capture developers already building on OpenAI-compatible agentic coding stacks.

DeepSeek V4 Pro 0813 arrives with a deliberate interoperability move: one OpenAI-compatible Responses API endpoint that lets existing Codex and OpenAI SDK stacks integrate without modification.

The API integration details matter. The example uses model deepseek-v4-flash, indicating DeepSeek's fast-tier model is the first or primary target for Responses API use, while the OpenRouter model page centers on deepseek-v4-pro-0813. The compatibility layer is deliberately permissive: unsupported parameters are silently ignored rather than rejected, and response object fields that depend on unsupported capabilities take fixed values, with store false, previous_response_id null, and parallel_tool_calls true. This means existing Responses API clients can connect without modification, but developers must understand the implications: no server-side persistence, no automatic prior-response linkage, and silent ignoring of parameters could mask unsupported behavior.

Streaming details also show real depth. The API uses server-sent events with a monotonically increasing sequence_number and terminates with response.completed, response.incomplete, or response.failed, not OpenAI's familiar data: [DONE] sentinel. This is a meaningful low-level difference for client libraries that scan for [DONE]; compatibility with the OpenAI SDK handles it, but custom clients need adjustment. For AI engineers, this is exactly the kind of edge-case that can break naive drop-in assumptions when bypassing the SDK.

OpenRouter's listing describes DeepSeek V4 Pro 0813 as a "large-scale mixture-of-experts model from DeepSeek" and "the GA release of DeepSeek V4 Pro." It is hosted by a single provider, so OpenRouter forwards every request directly with no routing decisions. The page tracks average price actually paid, throughput, latency, TTFT, and 30-day uptime, but the provided source excerpt does not include the actual numeric values. The presence of these metrics, and the "Apps" section for public apps sending the most traffic, signals that OpenRouter expects production workloads rather than experimental use. The activity section tracks token volume and request traffic over time. This is useful for adopters evaluating reliability and real-world performance.

What to Watch

Context is important here. DeepSeek has repeatedly disrupted the model market with capable open-weight models at low cost. V4 Pro 0813 as a GA release suggests the company is maturing its V4 line, with 0813 likely a dated checkpoint suffix. The simultaneous release of Responses API compatibility is not an accident: OpenAI's Codex CLI and API agents have become an important developer surface, and DeepSeek wants to be a drop-in alternative. For AI teams, this reduces migration friction and allows A/B testing between DeepSeek and OpenAI models with minimal code changes. However, the trade-off is that subtle differences in unsupported parameters, streaming completion events, and fixed fields require validation in production systems, especially for agents that rely on conversation state and tool-call parallelism.

Looking forward, expect DeepSeek to keep iterating on OpenAI compatibility, potentially expanding to more Responses API features, tool calls, and persistence. The model's MoE architecture could attract cost-sensitive builders if throughput and latency prove competitive. Enterprise AI teams will likely run internal evals comparing V4 Pro 0813 to prior DeepSeek versions and OpenAI or Anthropic models, focusing on coding, tool use, and long-horizon agent reliability. The single-provider hosting model on OpenRouter may limit redundancy, but DeepSeek's own endpoint provides a direct alternative. Overall, this is a tactical but consequential step in the ongoing API compatibility race, lowering switching costs and increasing pressure on foundation model pricing and developer tooling.

Timeline

Timeline

  1. OpenRouter lists DeepSeek V4 Pro 0813

  2. DeepSeek publishes Responses API guide

Source cluster

Primary reporting

2articles

Cite This Page

"DeepSeek V4 Pro 0813 GA: 1 endpoint for OpenAI Responses API drop-in." AI Intelligence Brief, August 12, 2026. https://getaibrief.com/story/deepseek-v4-pro-0813-ga-responses-api-ai

How we covered this story

Every story in our AI coverage is assembled from multiple primary sources, cross-referenced for factual consistency, and scored along three independent dimensions: sentiment, operational impact, and source-cluster confidence. Single-source rumors and unverifiable claims do not pass our editorial gate. When a story shows "Verified by N sources" with N≄2, the development is independently corroborated; when N=1, we mark it explicitly so readers can weigh the signal accordingly.

Impact scoring uses a 1-10 scale weighted toward regulatory, financial, and operational consequence rather than coverage volume. A topic that runs in every outlet but moves no real decisions ranks lower than a niche regulatory filing that reshapes how operators in the AI space have to behave. Read our full methodology for the scoring rubric, our glossary for term definitions, and our trends index for the longitudinal view across the beat.

Sources are only linked to a story once they clear our classification pipeline at a minimum 35 percent relevance threshold. According to that methodology, reviewed July 2026, this follows multi-source corroboration standards recommended by journalism research bodies such as the Reuters Institute for the Study of Journalism.

See something wrong in this story — a wrong fact, a broken source link, a misattributed entity? Report a data issue.