# DeepSeek V4 preview narrows the gap with open-weight MoE models
> V4-Pro at 1.6T params (49B active) and V4-Flash at 284B, both 1M-context, released open-weight on Hugging Face

**Meta:** type: event · date: 2026-04-24 · heads: اللعبة الطويلة, ما لا يقولونه · 6 takes · 3 lenses · 3 regions

## Summary

[Deepseek](/ar/entity/deepseek) released a preview of V4 on 24 April 2026, listing deepseek-v4-pro and deepseek-v4-flash as API model IDs and publishing open weights on Hugging Face. V4-Pro is a 1.6-trillion-parameter mixture-of-experts model with 49B activated; V4-Flash is 284B with 13B activated; both carry 1M-token context. DeepSeek claims V4 closes the gap with frontier models, beating some [GPT-5.2](/ar/entity/openai) and [Gemini 3.0 Pro](/ar/entity/google-deepmind) results on reasoning. It remains a preview with no finalisation date. The open-weight release, far cheaper per token, sharpens the [open-vs-closed split](/ar/n/open-vs-closed-frontier-2026) and runs against US [chip-export](/ar/n/h200-china-export-standoff) constraints DeepSeek trains under.

## By the numbers

- 1.6T, V4-Pro parameters (49B activated).
- 284B, V4-Flash parameters (13B activated).
- 1M tokens, context window, both models.
- 24 Apr 2026, preview release.
- Open weights, published on Hugging Face.

## Why it matters

A frontier-adjacent open-weight model from a Chinese lab undercuts the closed labs on price and removes the access gate the US is building around top US models, the asymmetry the [Anthropic export-control order](/ar/n/fable5-ai-export-controls) exposed. It keeps China competitive at the frontier despite chip controls.

## What to watch

- Independent benchmarks vs. GPT-5.5, Gemini 3.5 and Opus 4.8.
- A finalised (non-preview) V4 and pricing.
- Whether US export policy targets open-weight Chinese models.

## Regional takes (batched by bias / lens)

### unlabelled
- **DeepSeek API Docs (V4 preview)** (China, en) — DeepSeek's own release note announcing the V4 preview: deepseek-v4-flash and deepseek-v4-pro listed as available model IDs in the API; both MoE, 1M-token context, treated as a preview phase with no finalisation date.
  Source: https://api-docs.deepseek.com/news/news260424
- **Hugging Face (DeepSeek-V4-Pro)** (Global, en) — The open-weight model card: V4-Pro is a 1.6T-parameter MoE with 49B activated; V4-Flash 284B with 13B activated; both 1M-token context, the weights are downloadable.
  Source: https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro
- **SitePoint** (Global, en) — 
  Source: https://www.sitepoint.com/deepseek-v4-released-whats-new-in-the-latest-model-2026/
- **Evolink** (Global, en) — 
  Source: https://evolink.ai/blog/deepseek-v4-release-window-prep

### geopolitics / markets
- **CNBC** (United States, en) — Frames the V4 preview as China's open-source answer in an intensifying AI race, weighing it against US frontier labs and the chip-export constraints DeepSeek trains under.
  > "China's DeepSeek releases preview of its long-awaited V4 model as the AI race intensifies."
  Source: https://www.cnbc.com/2026/04/24/deepseek-v4-llm-preview-open-source-ai-competition-china.html

### capability
- **TechCrunch** (United States, en) — Reports DeepSeek's claim that V4 closes the gap with frontier models, outstripping some GPT-5.2 and Gemini 3.0 Pro results on reasoning tasks while remaining open-weight and far cheaper per token.
  > "DeepSeek previews a new AI model that 'closes the gap' with frontier models."
  Source: https://techcrunch.com/2026/04/24/deepseek-previews-new-ai-model-that-closes-the-gap-with-frontier-models/

## Across the graph
- Related: [[qwen-37-agents-open-vs-closed-2026]], [[open-vs-closed-frontier-2026]], [[h200-china-export-standoff]]
- Entities: Deepseek, China, Alibaba Qwen

---
Canonical: https://rbtfl.xyz/ar/n/deepseek-v4-open-weights-2026