TLDR: DeepSeek permanently cut DeepSeek V4 Pro pricing by 75 percent after extending a discount that was set to end May 31, 2026. The new rate runs $0.003625 to $0.87 per million tokens, reshaping AI agent and enterprise budgeting.
Key Takeaways:
- DeepSeek positions its agent models around cheap, long context, pitching an āeraā of 1M token practicality.
- DeepSeek V4 Pro now costs $0.003625 to $0.87 per one million tokens, down from $0.0145 to $3.48.
- Lower token prices could force competitors to respond or risk customers shifting compute budgets toward DeepSeek.
DeepSeek is turning pricing into a product feature, not a promo. If the math works for power users, other model makers may find discounting is no longer optional.
DeepSeek is turning pricing into a product feature, not a promo. If the math works for power users, other model makers may find discounting is no longer optional.
Q&A
If DeepSeek can sustain these per token prices, what does that suggest about its training and inference costs?
It likely reflects either improved efficiency at inference, cheaper compute supply, or tighter model optimization that reduces the cost per million tokens.
How could enterprise buyers react if DeepSeekās 1M context becomes the new default expectation?
Procurement teams may re benchmark vendor invoices around token throughput, prioritizing models that keep costs predictable as usage spikes.
What happens to rival pricing power if customers start treating context length as a budgeted line item?
Vendors may compete on unit economics rather than just quality, leading to more frequent price resets and stronger packaging around token budgets.
Why might competitors respond with targeted plans instead of matching the full discount immediately?
They could steer high volume users through negotiated tiers, impose usage limits, or bundle discounts that protect margins while still competing.
Could the earlier ādistillation attackā accusations accelerate legal or policy pressure on pricing and model access?
If scrutiny increases, rivals may push regulators on sourcing and training practices, while customers may demand transparency on model development and licensing terms.
No comments yet. Be the first to share your thoughts!