DeepSeek V4 Pro on OpenRouter: Model IDs, Pricing & Setup

OpenRouter added a fresh GA entry for DeepSeek V4 Pro on August 13, 2026, right when DeepSeek promoted the 0813 checkpoint to general availability^[1][2]. The listing exposes three V4-family IDs, and the pricing on the new entry matches DeepSeek's official flat rate^[1].

Two pro entries exist and it matters which one you pick. The old deepseek/deepseek-v4-pro entry dates from the April preview and still carries preview-era pricing ($1.168/$2.336) — nearly three times the GA price. The 0813 entry is the one to use^[1].

OpenRouter uses an OpenAI-compatible endpoint, so calling V4 Pro 0813 is a two-line change if you already use the official API: swap the base_url to openrouter.ai/api/v1 and the model to deepseek/deepseek-v4-pro-0813^[1].

• model: deepseek/deepseek-v4-pro-0813 (the GA entry, not the old preview one).
• reasoning_effort: low / high / max — same three levels as the official API^[3].
• thinking: enable with extra_body for chain-of-thought, matching DeepSeek's thinking mode^[3].

from openai import OpenAI

client = OpenAI(
    base_url="https://openrouter.ai/api/v1",
    api_key="<OPENROUTER_API_KEY>",
)

resp = client.chat.completions.create(
    model="deepseek/deepseek-v4-pro-0813",
    messages=[{"role": "user", "content": "Refactor this codebase summary"}],
    reasoning_effort="high",
    extra_body={"thinking": {"type": "enabled"}},
)
print(resp.choices[0].message.content)

On August 16 at 16:00 UTC, DeepSeek moved the official API to peak/off-peak pricing: Pro output goes from $0.87 to $1.98 off-peak or $3.96 at peak, with cache-hit input at $0.022/$0.044^[2][3]. As of this writing OpenRouter had not mirrored the new schedule — its 0813 entry still shows the flat $0.435/$0.87^[1].

That lag cuts both ways. If you route heavy Pro traffic through OpenRouter, you are currently paying the old flat rate — cheaper than DeepSeek's off-peak price only for cache-heavy workloads. If you are latency- or cost-sensitive to peak hours, OpenRouter's flat rate can be a hedge, but it may change without notice as OpenRouter reconciles pricing.

In practice, teams running OpenAI-SDK codebases switch to OpenRouter in minutes: same messages array, one base_url change, and the 0813 entry behaves like the official API including reasoning_content on thinking responses. The one difference to test first is streaming — OpenRouter adds its own SSE envelope — so verify your stream parser before migrating production traffic.

Cache pricing on OpenRouter matches DeepSeek's: $0.003625 per 1M cache-read tokens on the 0813 entry^[1]. Prefix caching is automatic on both sides.

OpenRouter supports bring-your-own-key (BYOK) for DeepSeek models, so you can pass a DeepSeek API key through OpenRouter and get direct billing from DeepSeek while using OpenRouter's unified SDK and fallback routing^[1]. This works on the 0813 entry like on any other DeepSeek listing.

If you primarily want the official API's rate limits and peak/off-peak pricing, go direct. If you want one SDK for many providers with automatic fallback, OpenRouter is the pragmatic middle ground^[1].

• BYOK: set an OpenRouter credentials object with your DeepSeek key; usage bills to DeepSeek.
• No BYOK: usage bills to OpenRouter credit at the listed rate.
• Concurrency on the official API is 500 for Pro / 2,500 for Flash^[3]; OpenRouter applies its own queueing on top.

The official API (base_url https://api.deepseek.com gives you the new peak/off-peak schedule, direct rate limits, and immediate access to new parameters like reasoning_effort without a middleman^[3]. OpenRouter gives you multi-provider routing, a single bill, and — right now — the older flat price^[1].

Most teams start direct on the official API, then add OpenRouter as a fallback or for provider-agnostic SDKs. Either way, the model string that matters for quality is deepseek-v4-pro-0813 — the GA checkpoint that lifted Terminal-Bench 2.1 to 87.9^[5].