DeepSeek הופכת את ההנחה ב-V4-Pro/Flash לקבועה

DeepSeek's tweet was short — 'making our discount permanent' — but the implication is loud. V4-Pro at 1.6T total / 49B active parameters with a 1M context window is being marketed as rivaling top closed-source models, and V4-Flash (284B / 13B active) as a faster, cheaper workhorse. By making the previously promotional price the new permanent price, DeepSeek is committing to a structural cost advantage rather than a temporary one.

Context: Alibaba's Qwen 3.7-Max just moved to paid-API-only for its frontier tier. Google's Gemini 3.5 Flash is also competing aggressively on price. DeepSeek is essentially calling the bottom of the price war and saying: this is the floor, plan around it. For builders running large agentic workloads — where 35-hour Qwen runs and Claude Code-style coding loops can rack up tokens fast — the permanent discount is the most consequential pricing news of the week.

What to watch: whether OpenAI or Anthropic match on Flash/Haiku-tier pricing post-IPO and post-Karpathy, or whether they let DeepSeek and Gemini Flash own the low end while they hold premium margins on frontier reasoning models.