DeepSeek releases V4-Flash with major post-training gains and open weights

DeepSeek launched the DeepSeek-V4-Flash API in public beta and immediately released its weights under an MIT license. The update delivers significant performance improvements over the April preview without altering the model's architecture or size, positioning it as a cost-effective alternative to proprietary models.

• The Flash API surpasses V4-Pro-Preview in agent capabilities and supports the Responses API format.
• Terminal-Bench scores jumped from 56.9 to 82.7, while GDPval-AA Elo rose from 1189 to 1559.
• The model retains its 284B total / 13B active parameter count and 1M context window.
• Pricing is set at $0.14/$0.28 per 1M tokens, with a 98% cache-hit discount reducing cached costs to $0.0028/1M.
• Open weights are available on Hugging Face, supported by vLLM and quantized deployments requiring roughly 168GB RAM for lossless 4-bit.

The release reframes the current price war by offering performance close to higher-end proprietary systems at a fraction of the cost, encouraging developers to integrate it into existing coding stacks.