DeepSeek Cuts V4-Pro Pricing by 75%

DeepSeek has attached a 75% discount promotion to its new V4-Pro input token pricing, running through May 5, 2026, and has immediately cut cache-hit costs across its API to one-tenth of the previous rate.

Even before the discount, V4-Pro was priced at $0.145/million tokens for input and $3.48/million tokens for output, making it cheaper than GPT-5.5, Gemini 3.1 Pro, and Claude Opus 4.7. With the promotion applied, the input price drops to roughly $0.036/million tokens.

• The smaller Flash variant is priced at $0.14 for input and $0.28/million tokens for output, undercutting GPT-5.4 Nano, Gemini 3.1 Flash, GPT-5.4 Mini, and Claude Haiku 4.5.
• V4-Pro is a Mixture-of-Experts model with 1.6 trillion total parameters and 49 billion active parameters, making it the largest currently released open-weight model.
• With a 1 million token context and Hybrid Attention Architecture, it targets long documents and large codebases, and it was trained and optimized for Huawei Ascend 950 and Cambricon hardware, reducing dependence on Nvidia.
• It also natively integrates with Claude Code, OpenClaw, and OpenCode, lowering the switching cost within existing agentic coding ecosystems.

The price war that DeepSeek ignited with R1 in January 2025 is intensifying again with this V4-Pro price cut. The move comes right after Michael Kratsios, Director of the White House Office of Science and Technology Policy, criticized large-scale model distillation by Chinese companies — meaning DeepSeek responded not with a direct rebuttal but with a price war. By removing model access barriers through open-source releases and lowering deployment costs with cheap API pricing and a 1 million token context, DeepSeek is increasing the incentive to switch away from OpenAI, Anthropic, and Google APIs.