Skip to content

Type: AI model

DeepSeek V4 Pro is the high-capability variant of DeepSeek’s V4 series previewed April 2026, a mixture-of-experts model with about 1.6T total and 49B active parameters and a 1-million-token context window. It is positioned near the frontier at low cost with open weights. GROUNDING tracks DeepSeek V4 Pro’s benchmarks, pricing, and availability.

Recent Updates

  • 2026-08-13: DeepSeek V4 Pro official release claims state-of-the-art performance on coding and agent benchmarks (量子位) · qbitai.comDeepSeek · Anthropic · OpenAI · Opus 4.8 · Fable 5 · DeepSeek-V4-Flash · GLM-5.2 · Kimi K3
  • 2026-08-13: Arena ranks DeepSeek V4 Pro as frontier-level, raises questions about AutoEval bias in model benchmarking (AI Projects) · t.meLLM Evals OpenAI · Anthropic · Arena · Claude · Qwen · GLM
  • 2026-08-14: DeepSeek-V4-Pro General Availability with Flexible Reasoning and Peak/Off-Peak Pricing (Hacker News) · api-docs.deepseek.comAgents DeepSeek · OpenAI · V4 Flash
  • 2026-08-17: Weekly AI digest: Qwen 3.8, Muse, Grok 4.6, Gemini Flash, GLM updates, and Cursor acquisition (эйай ньюз) · t.meOpen Source LLMs Alibaba · Meta · SpaceX · OpenAI · Google · Zhipu · DeepSeek · Qwen 3.8 27B · Qwen 3.8 Max · Muse Glimmer · Muse Spark 1.2 · Grok 4.6 · Gemini 3.7 Flash · GLM-5.3 · Sonnet 5
  • 2026-08-18: When Agentic Executions Fail: Detecting and Localizing Runtime Faults from Telemetry (cs.AI updates on arXiv.org) · arxiv.orgAgents Tool Use MCP
  • 2026-08-20: Building a Custom Watch Face with Claude on a $27 Smart Watch (Hacker News) · mikekasberg.comCode Agents Agents Open Source LLMs Steve Ruiz levelsio Claude · Kimi K3 · kimi-k2.6 · DeepSeek-V4-Flash · Fable
  • 2026-08-20: Every Model Cheats: Prompt-Level Mitigation of Cheating on Offensive Cyber Tasks (Hacker News) · dreadnode.ioLLM Evals Anthropic · OpenAI · Google · xAI · DeepSeek · Alibaba · z.ai · NIST HackTheBox · e2b Michael Kouremetis · Claude Opus 4.8 · Claude Opus 4.7 · Claude Opus 4.6 · Claude Sonnet 5 · Claude Sonnet 4.6 · Claude Haiku 4.5 · GPT 5.5 · GPT-5.4 · GPT-5.4-mini · Gemini-3.1-Pro · Gemini 3 Flash · Grok 4.20 · Grok 4.3 · DeepSeek-R1-0528 · DeepSeek-V4-Flash Qwen 3-7 Max Qwen 3.6 Max · Qwen 3.6 Plus Qwen3-Coder-Next · GLM-5.1 GLM-5-turbo
  • 2026-08-22: Frontier Model Autonomous Agent Benchmark: NanoGPT Optimizer Speedrun Leaderboard (breakingnewsofficial) · primeintellect.aiAgents Code Agents LLM Evals Anthropic · OpenAI · Moonshot AI · xAI · Alibaba · Zhipu · DeepSeek · Muse · Fable 5 · Opus 5 · Kimi K3 · Opus 4.8 · GPT-5.6 Sol GPT-5.6 Sol Pro · Sonnet 5 · GPT-5.6 Luna · Grok 4.5 · Qwen 3.8 Max · GLM-5.2 · GPT-5.6 Terra · Grok 4.6 · Muse Spark 1.2 · Muse Spark 1.1 · GPT 5.5 · Kimi K2.7
  • 2026-08-28: Automated exploit generation from rumored vulnerabilities breaks security embargo timelines (breakingnewsofficial) · anil.recoil.orgAgents Open Source LLMs Jane Street · Anthropic · DeepSeek · marimo · Langflow · Claude Fable · GPT-4
  • 2026-08-28: AI Coding Agents Finding Security Exploits Within Minutes of Patch Disclosure (breakingnewsofficial) · simonwillison.netAgents Code Agents Anthropic · DeepSeek · GitHub Simon Willison Anil Madhavapeddy Nick Craig-Wood · Claude Fable
  • 2026-09-03: 27B local model competes with trillion-parameter giants on agentic benchmarks through post-training (breakingnewsofficial) · qbitai.comAgents Tool Use StartLux DeepSeek · Qwen · Meta · Google · NVIDIA Chen Danian Guo Quanwei Luo Yongxiang Liang Wenfeng StartLux-V1.0-27B-Preview · DeepSeek-V4-Flash-0731 Step-3.7-Flash · Qwen 3.6-27B · Claude Sonnet 4.6
  • 2026-09-04: Reasoning Traces as State for Better Long-Context Encoding (breakingnewsofficial) · Z.AILong Context Context Engineering

Incident radar

GROUNDING’s Hallucination Incident Index has flagged 1 incident naming DeepSeek V4 Pro — a heuristic proxy over the AI-news radar’s daily journal (hallucination/jailbreak/refusal/bias keyword matches), not a verified incident registry.

  • Severity: S2 1
  • Category: Bias 1

Most recent:

  • Arena ranks DeepSeek V4 Pro as frontier-level, raises questions about AutoEval bias in model benchmarking source (2026-08-13)

Full breakdown: Hallucination Incident Index · Subscribe: incident-index RSS feed

FAQ

What is DeepSeek V4 Pro?

DeepSeek V4 Pro is the high-capability variant of DeepSeek’s V4 series previewed April 2026, a mixture-of-experts model with about 1.6T total and 49B active parameters and a 1-million-token context window. It is positioned near the frontier at low cost with open weights. GROUNDING tracks DeepSeek V4 Pro’s benchmarks, pricing, and availability.

What does this page track?

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

When was DeepSeek V4 Pro last mentioned?

DeepSeek V4 Pro was most recently mentioned in a radar update dated 2026-09-04.