DeepSeek V4 Series Debuts on Multiple Platforms with 1M Context
From Aug 15-17, DeepSeek V4 models launched on multiple hosting platforms: Cloudflare Workers AI added V4 Flash and Pro, while DeepInfra, Novita, and Together AI rolled out V4 Pro (0813). The common selling point is a 1M token (1,048,576) context window, suited for long-horizon agent workflows. V4 Pro uses a 1.6T MoE architecture with 49B active parameters and offers adjustable reasoning intensity (low/medium/high) for coding, agentic, and complex reasoning tasks. This series is seen as a significant step in long-context and efficient reasoning models.
Confirmed
- Cloudflare Workers AI hosts DeepSeek V4 Flash and Pro with full 1,048,576 token context (m1, m2).
- DeepInfra hosts DeepSeek-V4-Pro-0813 with 1.6T MoE, 49B active, 1M context, running on NVIDIA Blackwell in the US (m4).
- Novita supports DeepSeek-V4-Pro-0813 on Hugging Face with 1M context for reasoning, coding, and agents (m5).
- Together AI hosts DeepSeek V4 Pro 0813 with 1.6T total params (49B active), 1.05M context, and three reasoning modes (m3, m6).
Unconfirmed
- Context window specs vary slightly across platforms (1M vs 1.05M); official docs should be consulted.
- Whether the model is officially released by DeepSeek or third-party hosted remains unconfirmed.
Why it matters
DeepSeek V4's million-token context and efficient MoE architecture provide new infrastructure options for long-horizon agents and complex reasoning. Multi-platform availability suggests rapid ecosystem adoption, potentially driving broader use of ultra-long-context models.
2026-08-15 ~ 2026-08-17 · 6 related posts
Primary sources
- DeepSeek V4 on Cloudflare Workers AI: First 1M Token Context Models — ritakozlov ·
- Cloudflare Workers AI Launches DeepSeek V4 with 1M Context Window — michellechen ·
- DeepSeek V4 Pro launches on Together AI with 1M context — togethercompute ·
- [source] Cloudflare Workers AI Launches DeepSeek V4 with 1M Context Window — michellechen · 2026-08-15
- [source] DeepSeek V4 on Cloudflare Workers AI: First 1M Token Context Models — ritakozlov · 2026-08-15
- DeepSeek-V4-Pro-0813 Launches on Novita with 1M Context Window — huggingface · 2026-08-15
- DeepSeek-V4-Pro Launches on DeepInfra with 1.6T MoE and 1M Context — gharik · 2026-08-15
- [source] DeepSeek V4 Pro launches on Together AI with 1M context — togethercompute · 2026-08-17
1 near-duplicate retellings: togethercompute