DeepSeek’s New AI Model Is Reportedly 100 Times Cheaper to Run Than Claude Fable 5

Artificial intelligence may be advancing rapidly, but the next major disruption could come from something much less glamorous than machine consciousness:
Price.
DeepSeek has released V4 Flash, a lower-cost version of its flagship model that a research firm estimates is more than 100 times cheaper to run than Anthropic’s Claude Fable 5 on comparable benchmark workloads. �
Reuters
DeepSeek became internationally famous in early 2025 by challenging the assumption that competitive artificial intelligence required the same enormous spending associated with leading American companies.
V4 Flash returns to that formula: offer serious AI capability at a cost difficult for rivals to ignore.
How Cheap Is It?
DeepSeek lists V4 Flash at:
$0.14 per million input tokens
$0.28 per million output tokens
Artificial Analysis estimated that the average benchmark test cost roughly:
$0.03 with DeepSeek V4 Flash
$0.86 with Kimi K3
$1.86 with GPT-5.6 Sol
$3.15 with Claude Fable 5
That made V4 Flash approximately 105 times cheaper than Claude Fable 5 under the research firm’s testing method. �
Reuters
Is It Equally Intelligent?
No.
Lower price does not mean identical performance.
V4 Flash scored 50 on Artificial Analysis’s Intelligence Index, placing it alongside Google’s Gemini 3.6 Flash but behind stronger systems from OpenAI, Anthropic and Moonshot AI.
Kimi K3 scored 57, while the leading Anthropic and OpenAI systems were at least nine points higher than V4 Flash. �
Reuters
The important business question is therefore not:
“Is DeepSeek the world’s smartest model?”
It is:
“How many business tasks genuinely require the world’s smartest model?”
Why Cheap AI Can Be More Disruptive Than Powerful AI
A company may not need the most advanced reasoning system to:
Categorise customer messages
Summarise ordinary documents
Translate product descriptions
Draft routine emails
Extract invoice information
Produce standard reports
Moderate content
Power basic support bots
When millions of such tasks are performed every month, small differences in price become significant.
A premium model may produce a slightly better answer, but the cheaper model could make an entire business process financially viable.
Why the Story Went Viral
The headline challenges the economics underpinning the AI boom.
American companies are spending extraordinary amounts on chips, data centres, energy and model development.
DeepSeek’s strategy raises an uncomfortable question:
What happens when competitors offer adequate intelligence for a tiny fraction of the price?
The result could resemble earlier technology markets in which affordable alternatives expanded access and placed pressure on premium providers to justify their margins.
The Limitations Businesses Must Consider
Low cost should not be the only criterion.
Organisations must also evaluate:
Accuracy
Reliability
Data protection
Jurisdiction
Security
Availability
Language performance
Compliance requirements
Integration costs
Vendor stability
A model that saves money per token may become expensive if employees must repeatedly correct its mistakes.
The best measure is therefore cost per successful task, not merely cost per million tokens.
Mary Chuks’ Perspective
This is where practical AI separates itself from AI fandom.
Businesses should not select models because the internet has declared one company the winner.
They should create a workload map.
Use frontier models for high-value reasoning, complex research and decisions where errors are expensive.
Use efficient lower-cost models for repetitive tasks with clear verification.
That is how AI becomes a scaler rather than an uncontrolled monthly bill.
Conclusion
DeepSeek V4 Flash does not need to defeat every premium AI system to transform the market.
It needs to become good enough for a large percentage of ordinary business work.
The next phase of AI competition may not ask who created the smartest model.
It may ask who delivered the most useful intelligence for three cents.
Original Sources and Further Reading
Original report: Reuters, Eduardo Baptista, “DeepSeek’s New AI Model Is by Far the Cheapest of Well-Known Models to Run,” published 3 August 2026. �
Reuters