DeepSeek V4 Flash 0731 ships on Together AI with 284B parameters and 1M context
togethercompute · x · 2026-08-04
Together AI’s model page for DeepSeek V4 Flash 0731 adds concrete details to the launch:
- 284B total parameters with 13B active per token
- 1M-token context window on a hybrid attention design
- three reasoning-effort levels: low, high, and max
- a focus on agentic coding and long-context production use
- released under the MIT license and offered on Together AI infrastructure
Together says the model is aimed at terminal, repository, and security-engineering tasks, and claims it outperforms the larger V4 Pro preview on its published agentic benchmark set despite using far fewer active parameters.
Related event: Together AI Launches DeepSeek V4 Flash(3 posts)→
More from coding & agent
- Claude Fable 5 helps ship a SNES puzzle game that runs on real hardware — AIandDesign · 2026-08-04
- Opus 5 and Gauntlet Loop Ship a Web-Native Game in Days — mattshumer_ · 2026-08-04
- Creator claims 47 n8n agents built with Claude, shares 3 prompts that actually work — Aiden_Tech_Ai · 2026-08-04
- Dorsa Rohani refreshes her personal site with multi-agent AI and open-source projects — dorsa_rohani · 2026-08-04
- Claude 3.5 Sonnet Aces Adversarial Data Science Test, Catches Data Leakage Autonomously — hugobowne · 2026-08-04
- opencode v1.18.12 fixes Azure GPT-5.5+ reasoning failures and desktop lag — opencode-agent[bot] · 2026-08-04