DeepSeek's official release of the V4 preview large language model on April 24, 2026, has driven trader consensus to a near-certain 100% implied probability for "on or prior to April 24," as the open-source weights for DeepSeek-V4-Pro (1.6T total parameters via mixture-of-experts architecture, 49B active) and V4-Flash (284B total, 13B active) went live on Hugging Face alongside a detailed technical report touting 1M-token context length and frontier-level benchmarks in coding and math. This follows months of delays tied to Huawei Ascend chip optimization, validating prior leaks and positioning DeepSeek competitively against closed models like Gemini 3.1 Pro. Resolution risks remain minimal barring disputes over "preview" status, with full V4 potentially imminent.