DeepSeek V4 Pro Confirmed: The 1.6T Open-Weight Flagship

DeepSeek’s August is the story nobody saw coming. Within a week of its V4-Flash 0731 retrain — a sparse mixture-of-experts model at 284 billion parameters that cut the gap to frontier closed models to single digits on several leaderboards — the company confirmed an official V4 Pro is coming: a 1.6 trillion-parameter open-weight flagship, with no date attached.

The market reaction was visible: the open-weight release moved Nasdaq-listed AI infrastructure names, and Western analysts spent the week re-baselining what “frontier” means when a Beijing lab ships weights, not API endpoints.

Key facts

• V4 Pro confirmed by DeepSeek itself — a 1.6T-parameter flagship with open weights; no release date announced.
• V4-Flash 0731 (released Jul 31–Aug 1) — the 284B-parameter sparse MoE retrain that reset the open-vs-closed economics conversation; closes the US-China gap to single digits on key evals per third-party ranking.
• Market effect: the release week moved Nasdaq AI names and pulled follow-on price cuts across closed APIs (OpenAI’s Luna −80% on Jul 30 sits in the same arc).
• Positioning: DeepSeek remains the pricing anchor of the open-weight era — per-test economics around three cents, open licensing, and a publicly scheduled path to a 1.6T flagship.
• Skeptics’ note: V4-Flash-0731’s benchmark claims are self-reported by DeepSeek; independent replication (Artificial Analysis, LMArena retests) was still converging at time of writing.

Why it matters

• The open frontier is being scheduled, not improvised. V4-Flash 0731 now, V4 Pro later — DeepSeek is the only lab publishing a roadmap that top-tier closed vendors have to price against even before it ships.
• 1.6T open weights would be the largest ever released. If it matches V4-Flash’s cost-per-test discipline, the entire premium segment of closed APIs comes under structural pressure.
• The gap conversation flipped: it’s no longer “China vs US”; it’s “open weights vs closed APIs” with the US-China axis folded into it — and open just gained single-digit ground.

What to watch

• The V4 Pro date and context window — the two specs that determine whether the release is a headline or a paradigm event.
• Independent retests of V4-Flash 0731 — trusted numbers will set the baseline expectation for Pro.
• Closed-lab counter-moves — OpenAI and Anthropic have already answered July with price cuts; a 1.6T open release invites a second round.

Official source

• DeepSeek: V4-Flash release notes and API documentation
• Renovate QR (verified Aug 9): Best AI Models August 2026 — DeepSeek V4 Pro confirmed
• Release timeline: LLM Market Cap — DeepSeek V4 Flash series

Updated August 10, 2026.