DeepSeek V4 Pro Release Date: From Preview to GA (2026)

DeepSeek V4 Pro went generally available on August 13, 2026, when DeepSeek shipped the **0813 checkpoint** — the build that moved the 1.6T-parameter flagship from preview to production^[1]. The GA announcement is dated 2026/08/13 in the official API changelog, and every service (API, app, and web) switched to the 0813 build the same day.

That date ends a nearly four-month preview: V4 Pro launched as a preview on April 24, 2026, and every build between April and August was explicitly labeled a preview. The 0813 release is the first production build, and the model ID on the API stayed `deepseek-v4-pro` — no migration needed for existing callers^[3].

The August 13 launch landed in the same window as another DeepSeek release: the DeepSeek Harness developer preview, an MIT-licensed agent framework that DeepSeek used to produce the official agent benchmarks for the 0813 build^[1].

• GA date: August 13, 2026 (0813 checkpoint)
• Preview period: April 24, 2026 – August 13, 2026
• Model ID unchanged: deepseek-v4-pro
• Announced alongside the DeepSeek Harness developer preview

The path from preview to GA had several milestones, and most of them changed what you pay and how you call the model.

The pre-GA rumor mill was loud but mostly wrong on timing: mid-July windows and 'as early as tomorrow' claims all missed, and the actual GA landed on August 13. The reliable signals were official — the July 31 changelog explicitly said the Pro release 'will follow soon,' which pointed at early-to-mid August^[1].

For context on the smaller sibling released the same day as the preview, see the DeepSeek V4 Flash guide; Flash hit its own official build (Flash-0731) on July 31, two weeks before Pro went GA.

The GA release was not just a status flip — it shipped three headline upgrades: major agent improvements, flexible reasoning effort, and native OpenAI Responses API support^[1].

On benchmarks, the 0813 build is a large jump over the preview: Terminal Bench 2.1 goes from 72.1 (preview) to 87.9, and the build takes first place on Cybergym (83.3) and AutomationBench (31.8) in DeepSeek's official tables^[4]. Independent testing by MindStudio scored it 76.25% on an 8-task coding/reasoning suite, versus 24.8% for the preview[6].

Those official agent numbers were produced with the DeepSeek Harness in minimal mode at max reasoning effort — see the DeepSeek Harness benchmarks page for how the harness affects the scores.

• Major agent upgrades with strong production gains (DeepSeek's phrasing)
• Flexible reasoning effort for V4 Pro and V4 Flash: low / high / max
• Native OpenAI Responses API support, optimized for Codex with one-click setup
• V4 Pro available on app and web through 'Expert Mode'
• Peak and off-peak API pricing introduced (off-peak 50% lower than peak)

You do not need to do anything special to move to GA — the model ID is unchanged and the API already points at the 0813 build^[3].

• API: model deepseek-v4-pro, base URL https://api.deepseek.com
• App/web: switch the model menu to V4 Pro — it runs in 'Expert Mode'
• Anthropic-format callers: use https://api.deepseek.com/anthropic
• Codex users: the official one-click setup script configures Responses API support

from openai import OpenAI

client = OpenAI(
    api_key="<DeepSeek API Key>",
    base_url="https://api.deepseek.com",
)

response = client.chat.completions.create(
    model="deepseek-v4-pro",
    messages=[{"role": "user", "content": "Explain the agent upgrade in the 0813 build."}],
    reasoning_effort="high",  # low / high / max
)

The biggest post-GA change is pricing: peak/off-peak rates took effect at 16:00 UTC on August 16, 2026 — three days after GA^[1][2].

Under the new schedule, V4 Pro costs $1.98 per 1M output tokens off-peak and $3.96 peak, up from $0.87 flat. If you are planning capacity, the full V4 Pro pricing page has the complete table and the peak vs off-peak scheduling guide explains how to route workloads to the cheap hours.

Beyond pricing, watch for the ecosystem around the model: the Responses API setup page covers the Codex one-click config, and the 0813 build explainer details what changed versus the preview build.