Yesterday in AI: 24 April 2026 — DeepSeek V4 Rivals Frontier at 1/6th the Price

DeepSeek drops V4 (1.6T params, MIT license, 80.6% SWE-bench) at one-sixth the cost of Opus 4.7; Tencent open-sources Hy3 Preview; and OpenAI launches always-on Workspace Agents to replace Custom GPTs.

By OMC Editorial on 2026-04-25

DeepSeek V4: Near-Frontier Intelligence at One-Sixth the Cost One year after rattling Silicon Valley, DeepSeek released two open-weight models under the MIT license on April 24: V4-Pro 1.6T parameters, 49B active and V4-Flash 284B total, 13B active. Both support a 1-million-token context window, trained on 32 trillion tokens. A new Hybrid Attention architecture combining Compressed Sparse Attention and Heavily Compressed Attention reduces inference FLOPs by 73% versus V3 at 1M-context — meaning the cost advantage isn't just pricing strategy, it's baked into the architecture. V4-Pro scores 80.6% on SWE-bench Verified, just 0.2 points behind Claude Opus 4.6 Max 80.8%. API pricing for Pro sits at $1.74/$3.48 per million tokens input/output — roughly one-sixth the cost of Opus 4.7. Weights are MIT-licensed and available at deepseek-ai/DeepSeek-V4-Prohttps://huggingface.co/deepseek-ai/DeepSeek-V4-Pro and deepseek-ai/DeepSeek-V4-Flashhttps://huggingface.co/deepseek-ai/DeepSeek-V4-Flash on Hugging Face. Sources: VentureBeathttps://venturebeat.com/technology/deepseek-v4-arrives-with-near-state-of-the-art-intelligence-at-1-6th-the-cost-of-opus-4-7-gpt-5-5 | TechCrunchhttps://techcrunch.com/2026/04/24/deepseek-previews-new-ai-model-that-closes-the-gap-with-frontier-models/ | Simon Willisonhttps://simonwillison.net/2026/Apr/24/deepseek-v4/ --- Tencent Open-Sources Hy3 Preview: China's New Agentic Contender Tencent's Hunyuan team open-sourced Hy3 Preview on April 23-24, the first major model release since the company restructured its AI division under new leadership. Hy3 is a 295B-parameter MoE 21B active, 256K context designed specifically for reasoning and long-horizon agentic tasks. Tencent claims a 40% efficiency gain over its predecessor and has already deployed Hy3 in Yuanbao, its flagship chatbot, replacing DeepSeek as the primary underlying model — a telling sign of how quickly this tier of Chinese open-source has matured. Weights are available at tencent/Hy3-p