DeepSeek V4 Pro launches on Together AI with 1.6T MoE and 1M context
togethercompute · x · 2026-08-17
DeepSeek V4 Pro 0813 is officially released and available on Together AI. The model features a Mixture-of-Experts (MoE) architecture with 1.6T total parameters (49B active per token) and a 1.05M token context window.
Key capabilities include:
- Adjustable Reasoning Effort: Offers Low, High, and Max levels to trade speed for reasoning depth per request.
- Agentic Focus: Significant gains in production agentic tasks, particularly in repository-scale engineering and multi-tool agent tasks.
- Efficient Inference: Includes a DSpark speculative decoding module for faster generation.
- Open License: Released under the MIT license.
Benchmarks show improvements across the board, including Fronti erMath, GPQA Diamond, and Terminal-Bench, superseding the previous preview version.
Related event: DeepSeek V4 Series Debuts on Multiple Platforms with 1M Context(6 posts)→
More from Models
- Why aren't labs training on hundreds of millions of videos for native audio/video? — imjustnewatai · 2026-10-16
- ChatGPT UI screenshot reveals Wallet coming soon plus Email, Calendar, Health and Finances integrations — jxnlco · 2026-10-16
- Why the next OpenAI and Anthropic models may still disappoint: no native audio or video input — imjustnewatai · 2026-10-16
- Thousands Hit by ChatGPT Standard Voice Mode Regression: Wrong Voice, Interruptions — shaunralston · 2026-10-16
- Gemini Argon 4 Quota Exhausted, Reset Countdown Shows Nearly 7 Days — ssh4net · 2026-10-16
- AINews: Meta ships 816B Muse Spark Open; Goodfire argues neural nets are already clean — Latent Space · 2026-10-16