Skip to content

Type: AI model

DeepSeek V4 is DeepSeek’s frontier mixture-of-experts model series previewed April 2026, offered in Pro and Flash variants with up to a 1-million-token context window and open weights under an MIT license. It is positioned near the frontier at a fraction of competitors’ cost. GROUNDING tracks DeepSeek V4’s variants, pricing, and benchmarks.

Recent Updates

  • 2026-06-28: DeepSeek claims DSpark improves V4 serving throughput with speculative decoding (alphaXiv) · twitter.comalphaXiv · DeepSeek · X
  • 2026-06-29: DeepSeek V4 is slated for mid-July with peak-hour API pricing (Hacker News) · kucoin.comDeepSeek ME News BlockBeats KuCoinFlash · Hacker News · DeepSeek V4 Pro · DeepSeek V4 Flash
  • 2026-06-30: DeepSeek V4 official version will raise peak-hour API prices and claims capability upgrades (量子位) · qbitai.comLong Context Agents DeepSeek 0xSupergemma · QbitAI · 量子位 东数西算 衡宇 Jun Song 陈德里 唐杰 库克 V4 V3 · R1
  • 2026-06-30: Enterprises are shifting AI workloads toward cheaper open-weight models to cut token costs (Все статьи подряд / Искусственный интеллект / Хабр) · habr.comOpen Source LLMs Coinbase · Snowflake Lindy · OpenAI · Microsoft · NVIDIA Broadcom · Azure Azure Local · vLLM MLSys Brian Armstrong · gpt-5.5-thinking · Opus 4.8 · GLM-5.2 Kimi 2.7
  • 2026-07-04: 35 RL interview questions compiled from Zhihu and recent discussions (Все статьи подряд / Искусственный интеллект / Хабр) · habr.comLLM Evals DeepSeek R1 · DeepSeek-V3.2
  • 2026-07-06: OpenSquilla 0.5.0 Preview adds multi-model agent routing and tops DRACO benchmark tables (量子位) · qbitai.comAgents Tool Use LLM Evals OpenSquilla TokenRhythm · DeepSeek · GLM · Kimi · Qwen · Anthropic · Brave Search · DuckDuckGo · GLM-5.2 · Kimi K2.7 · Qwen3.7 · Opus 4.8 · GPT 5.5 · Fable 5
  • 2026-07-06: OpenSquilla 0.5.0 Preview adds multi-model agentic routing and reports DRACO leaderboard wins (量子位) · qbitai.comAgents Tool Use LLM Evals OpenSquilla TokenRhythm · DeepSeek · GLM · Kimi · Qwen · Anthropic · Brave Search · DuckDuckGo · DRACO · Quantum Bit · GLM-5.2 · Kimi K2.7 · Qwen3.7 · Opus 4.8 · GPT 5.5 · Fable 5
  • 2026-07-06: ms-swift ranks #4 on GitHub AI with unified fine-tuning support for 600+ text and 400+ multimodal models (GitHub AI Ranking Changes (Top 10)) · github.comOpen Source LLMs ModelScope · Qwen3 · Qwen3.5 · Qwen3.6 · DeepSeek R1 DeepSeek-VL2 Llama4 InternLM3 GLM4.5 GLM4.5-V · GLM-5.1 · Mistral · Qwen3-VL Qwen3-Omni InternVL3.5 Ovis2.5 · Gemma4 · LLaVA Phi4 MiniCPM-V-4
  • 2026-07-07: DSpark proposes confidence-scheduled speculative decoding for faster LLM serving (cs.AI updates on arXiv.org) · arxiv.orgContext Engineering DeepSeek MTP-1
  • 2026-07-07: How scam-style prompts tested seven top LLMs (Все статьи подряд / Искусственный интеллект / Хабр) · habr.comLLM Evals CyberOK Research Sergey Gordeychik Claude Opus 4.8 · Claude Sonnet 4.6 · GPT 5.5 Qwen 3.7 · Mistral Llama-8B
  • 2026-07-07: Why inline keywords and code docs can help sparse attention (AI Projects) · t.meContext Engineering Embeddings DeepSeek · GigaChat
  • 2026-07-13: Control the ideas, not the code (Hacker News) · antirez.comRedis · DeepSeek · X · GLM-5.2
  • 2026-07-13: Inspur positions CPU-heavy liquid-cooled racks and multi-model fusion for the agent era (量子位) · qbitai.comAgents Tool Use LLM Evals Inspur Information IDC · Gartner · QbitAI YuanNao EPAI YuanNao SD200 Cresi Zhao Shuai · kimi-k2.6 · GLM-5.2 · MiniMax M3
  • 2026-07-15: Anthropic Research: Four New Agentic Misalignment Failure Modes in Frontier Models (Anthropic) · alignment.anthropic.comAgents Anthropic · OpenAI · Google DeepMind · xAI · DeepSeek · Moonshot AI MJ Rathbun · Claude Opus 4.8 · Claude Opus 4.7 · Claude Opus 4.6 · Claude Opus 4.5 · Claude Sonnet 4.6 · Claude Mythos Preview · GPT 5.5 · GPT-5.4 · Gemini-3.1-Pro · Gemini 3 Flash · Gemini 3.5 Flash · Grok 4.3 · kimi-k2.6
  • 2026-07-16: Claude Code-style agent architectures and the cost of dropping context (AI Projects) · agents.mdAgents Context Engineering Anthropic · DeepSeek · Huawei GenAI
  • 2026-07-21: Perplexity as a signal of model similarity and ensemble diversity (@turboproject) · t.meLLM Evals DeepSeek · LM Arena · Kimi K3 · Fable 5 · Claude

FAQ

What is DeepSeek V4?

DeepSeek V4 is DeepSeek’s frontier mixture-of-experts model series previewed April 2026, offered in Pro and Flash variants with up to a 1-million-token context window and open weights under an MIT license. It is positioned near the frontier at a fraction of competitors’ cost. GROUNDING tracks DeepSeek V4’s variants, pricing, and benchmarks.

What does this page track?

Dated radar mentions, source links, related concepts, and builder-relevant context for DeepSeek V4, collected automatically by GROUNDING.

When was DeepSeek V4 last mentioned?

DeepSeek V4 was most recently mentioned in a radar update dated 2026-07-21.