DeepSeek V4-Flash-0731 Official Release — $0.14/M Input, 1M Context

🤖 AI Tool Alert — August 1, 2026

DeepSeek V4-Flash-0731 Official Release — $0.14/M Input, 1M Context

DeepSeek released V4-Flash-0731 on July 31 — official public beta of their V4-Flash line.

The specs:

• Architecture: Sparse MoE — 13B active params / 284B total


• Context window: 1 million tokens


• Pricing: $0.14/M input (cache miss), $0.0028/M (cache hit)


• Terminal-Bench: 82.7% — beats its own 1.6T-param V4-Pro on agent benchmarks


• Open weights on Hugging Face: deepseek-ai/DeepSeek-V4-Flash-0731

Why it matters:

For a codebase-aware agent loading 300K tokens of context:

• GPT-class model: ~$15-30 per run, 15-30s latency


• V4-Flash-0731: ~$0.05 per run, comparable or faster latency


• Savings on 1,000 runs: ~$14,950

Despite activating only 13B params per forward pass (vs 1.6T for V4-Pro), it outperforms the larger model on agent benchmarks. The MoE efficiency means faster responses without quality loss.

Concrete use case: Swap your model endpoint to deepseek-v4-flash-0731 for code review agents, automated testing, or PR summarization. The 1M context window handles entire repos, and the open weights let you self-host or fine-tune on your codebase patterns.

🔗 DeepSeek API | HF Weights