Qwen vs DeepSeek Pricing: API Costs, Self-Hosting & Total Cost Comparison

As organizations increasingly integrate AI into software development, customer support, enterprise search, and business automation, one question consistently rises to the top:

Which AI model delivers the best value for money?

For engineering leaders, startups, and enterprises, choosing between Qwen and DeepSeek isn't simply about benchmark scores or coding performance. The long-term cost of deploying and operating an AI model can significantly influence project success.

A model with lower API pricing may require more powerful hardware for self-hosting. Another model may cost slightly more per million tokens but reduce infrastructure expenses through higher efficiency or faster inference. These trade-offs mean that evaluating pricing requires a broader perspective than comparing token rates alone.

Today, both Qwen and DeepSeek offer flexible deployment options. Organizations can consume them through managed APIs, deploy them on cloud GPU infrastructure, or run them privately using open-source inference frameworks. Each approach has different cost implications depending on traffic volume, latency requirements, compliance needs, and operational complexity.

Questions that buyers frequently ask include:

• Is Qwen cheaper than DeepSeek?
• Which API has lower token costs?
• Is self-hosting more affordable than managed APIs?
• How much does it cost to run these models on AWS?
• Which model provides better ROI for startups?
• Which option is most cost-effective for enterprise AI?

These questions become even more important as AI usage scales from a few thousand requests per month to millions of API calls across multiple applications.

Rather than focusing solely on published pricing tables, this guide examines the Total Cost of Ownership (TCO) of both ecosystems. We'll compare API pricing, GPU infrastructure costs, deployment strategies, operational overhead, and optimization techniques to help organizations make informed financial decisions. For detailed guidance on deploying these models in production, see our guide on Deploy LLMs on AWS.

Whether you're building an AI coding assistant, an enterprise chatbot, a Retrieval-Augmented Generation (RAG) platform, or an internal AI agent, understanding the economics of Qwen and DeepSeek is just as important as understanding their technical capabilities.

What This Guide Covers

• API pricing models
• Token costs
• Long-context pricing
• Self-hosting expenses
• GPU requirements
• Cloud deployment costs
• Cost optimization strategies
• Infrastructure planning
• Enterprise ROI
• Total Cost of Ownership (TCO)

By the end of this comparison, you'll have a practical framework for selecting the most cost-effective AI solution based on your organization's workload, growth plans, and infrastructure strategy.

Why AI Pricing Matters

Many organizations begin evaluating AI models by comparing benchmark scores.

However, once a model moves into production, financial considerations quickly become just as important as technical performance.

For example, an AI coding assistant used by 20 developers has very different cost dynamics than a customer support platform processing millions of requests every month.

Even small differences in token pricing or infrastructure efficiency can result in substantial cost variations at scale.

AI pricing affects:

• Monthly operating expenses
• Product profitability
• Infrastructure planning
• Customer pricing strategies
• Return on investment (ROI)
• Scalability

As AI adoption grows, controlling inference costs becomes an essential component of long-term business planning.

The Four Components of AI Costs

Many buyers assume API pricing is the only expense associated with Large Language Models.

In reality, organizations should evaluate four major cost categories.

A model with lower token pricing may ultimately cost more if it requires additional engineering effort or expensive GPU infrastructure.

EaseCloud Insight

At EaseCloud, we encourage organizations to evaluate AI projects using Total Cost of Ownership (TCO) rather than API pricing alone. Hidden operational costs—such as infrastructure management, monitoring, autoscaling, and maintenance—often exceed the savings gained from choosing the cheapest API.

Understanding How LLM Pricing Works

Large Language Models are priced differently from traditional cloud services.

Instead of charging by CPU time or storage alone, providers typically bill based on tokens, which represent pieces of text processed by the model. For a comparison of managed ML platforms, see SageMaker vs Azure ML vs Vertex AI.

Every request consists of two primary components:

• Input tokens
• Output tokens

Some providers also charge separately for cached tokens or long-context processing.

Understanding these pricing models is essential before comparing Qwen and DeepSeek.

What Are Tokens?

Tokens are the units AI models use to process text.

A token may represent:

• A word
• Part of a word
• A punctuation mark
• A number
• A symbol

For example, a short paragraph may contain several hundred tokens, while a lengthy technical document can contain tens of thousands.

Every API request consumes tokens during:

• Prompt submission
• Context processing
• Response generation

Higher token usage directly increases operating costs.

Input Tokens vs Output Tokens

Most AI providers price these separately.

Input Tokens

Input tokens include everything sent to the model.

Examples:

• User prompts
• System prompts
• Previous conversation history
• Retrieved documents
• Programming code
• Knowledge base content

Applications with long prompts or extensive Retrieval-Augmented Generation (RAG) pipelines often consume a significant number of input tokens.

Output Tokens

Output tokens represent the text generated by the AI model.

Examples include:

• Generated code
• Chat responses
• Reports
• Summaries
• Documentation
• SQL queries

Applications that require long-form responses generally incur higher output token costs.

Why This Matters

Two models with identical input pricing may have very different output pricing, leading to noticeable differences in monthly expenses depending on the application's response length.

Cached Tokens

Many enterprise AI platforms optimize costs by reusing portions of previous prompts.

These reused sections are referred to as cached tokens.

Common examples include:

• Repeated system prompts
• Company policies
• Product documentation
• Shared instructions
• AI agent configurations

Because cached content doesn't require the same level of processing, some providers apply discounted pricing to these tokens.

For organizations with repetitive workflows, prompt caching can significantly reduce overall API expenses.

Context Window Pricing

Modern LLMs support increasingly large context windows, allowing them to process:

• Large code repositories
• Technical documentation
• Research papers
• Contracts
• Enterprise knowledge bases
• Financial reports

While longer context windows enable more sophisticated applications, they also increase the number of input tokens processed per request.

Organizations designing Retrieval-Augmented Generation (RAG) systems should balance context size with cost efficiency rather than assuming the largest context window is always the best option.

Qwen Pricing Overview

Qwen is available through multiple deployment models, giving organizations flexibility based on their technical and financial requirements. For deploying Qwen on AWS SageMaker with auto-scaling, see our guide on Deploy Qwen with SageMaker Auto-Scaling.

Common deployment options include:

• Managed APIs
• Alibaba Cloud Model Studio
• OpenRouter
• Together AI
• Fireworks AI
• Self-hosted deployments

Pricing varies depending on:

• Model variant
• Provider
• Region
• Context length
• Token volume
• Service-level agreements (SLAs)

This flexibility allows teams to choose between fully managed services and private infrastructure depending on workload requirements.

DeepSeek Pricing Overview

DeepSeek also supports a broad range of deployment strategies.

Organizations can access DeepSeek through:

• Managed APIs
• DeepSeek Platform
• OpenRouter
• Together AI
• Fireworks AI
• Self-hosted inference servers

DeepSeek has gained attention for offering highly competitive pricing while maintaining strong reasoning and coding capabilities.

However, as with Qwen, total cost depends on more than API rates alone. Infrastructure choices, inference optimization, and operational efficiency all contribute to long-term expenses.

Comparing Pricing Models

Although both ecosystems provide flexible pricing options, their cost structures are influenced by several common factors.

The most cost-effective option depends on:

• Monthly API usage
• Concurrent users
• Latency requirements
• Compliance obligations
• Cloud infrastructure
• Engineering resources

Organizations should evaluate these factors together rather than focusing on a single pricing metric.

How to Calculate the Total Cost of Ownership (TCO)

API pricing is only one part of an AI budget. A more accurate assessment considers all expenses associated with deploying and operating AI at scale.

A practical TCO framework includes:

For startups with modest traffic, managed APIs often provide the lowest upfront cost.

For enterprises processing millions of requests each month, self-hosting may reduce long-term expenses—provided the organization has the expertise to manage infrastructure effectively.

EaseCloud Cost Evaluation Framework

At EaseCloud, we assess AI pricing using five key dimensions before recommending a deployment model:

1. Workload Volume

We estimate expected API requests, token usage, and growth projections.

2. Infrastructure Strategy

We determine whether managed APIs, private GPU clusters, hybrid deployments, or multi-cloud architectures provide the best balance of cost and performance.

3. Performance Requirements

We evaluate latency, throughput, concurrency, and regional availability to ensure infrastructure aligns with business objectives.

4. Governance & Security

Compliance, data residency, and access controls often influence deployment decisions as much as pricing.

5. Long-Term ROI

Rather than minimizing short-term expenses, we focus on reducing the overall cost of operating AI over the lifecycle of the application.

This framework helps organizations avoid choosing an AI model based solely on published token prices while overlooking infrastructure complexity or operational overhead.

API Pricing Comparison

Organizations typically consume Qwen and DeepSeek through managed APIs before considering self-hosting.

Although providers use different pricing structures, most charge based on:

• Input tokens
• Output tokens
• Cached tokens (where supported)
• Context length
• Premium model variants
• Rate limits
• Throughput tiers

The actual price you pay depends not only on the model, but also on the provider you select.

Qwen API Pricing

Qwen models are available from multiple providers, including:

• Alibaba Cloud Model Studio
• OpenRouter
• Together AI
• Fireworks AI
• Hugging Face Inference Endpoints
• Community inference platforms

Pricing can differ because providers include:

• Different infrastructure
• Different GPUs
• Various service-level agreements
• Regional hosting
• Enterprise support

Advantages

• Multiple deployment choices
• Flexible pricing options
• Enterprise cloud ecosystem
• Strong regional availability
• Mature infrastructure

Considerations

• Costs vary by provider
• Premium models may have separate pricing
• Regional availability affects pricing

DeepSeek API Pricing

DeepSeek models are available through:

• DeepSeek Platform
• OpenRouter
• Together AI
• Fireworks AI
• Self-hosted inference

DeepSeek has gained significant attention because its APIs are often positioned as cost-effective while delivering strong reasoning performance.

Advantages

• Competitive pricing
• Strong reasoning models
• Excellent coding capability
• Open-weight deployment options

Considerations

• Provider pricing differs
• Enterprise SLAs vary
• Heavy demand may occasionally affect public endpoints

EaseCloud Insight

When evaluating AI APIs, EaseCloud recommends comparing providers rather than models alone. Two organizations using the same model may experience very different costs due to provider-specific pricing, throughput limits, and infrastructure optimizations.

Cost Per Million Tokens

Most AI providers express pricing as the cost per million input and output tokens.

Understanding token consumption is more useful than memorizing price tables because it allows organizations to estimate expenses across different workloads.

Typical workloads include:

Applications processing large code repositories or enterprise documents generally consume significantly more input tokens than conversational chatbots.

Long Context Pricing

One of the biggest cost drivers is the context window.

Modern applications increasingly process:

• Large codebases
• Technical documentation
• Research papers
• Contracts
• Knowledge bases
• Multi-file repositories

Larger context windows increase:

• Input tokens
• Memory usage
• GPU utilization
• Inference time

Organizations should avoid automatically sending the maximum context with every request.

Instead, implement intelligent retrieval strategies that provide only the most relevant information.

EaseCloud Recommendation

For Retrieval-Augmented Generation (RAG) systems, EaseCloud recommends optimizing document retrieval before increasing context size. Efficient retrieval pipelines often reduce token consumption while improving answer quality.

Batch Processing Costs

Organizations serving thousands of requests simultaneously often reduce costs through batch inference.

Benefits include:

• Higher GPU utilization
• Lower cost per request
• Improved throughput
• Better infrastructure efficiency

Batch inference is particularly valuable for:

• AI document processing
• Report generation
• Offline analytics
• Bulk summarization
• Code analysis

Inference frameworks such as vLLM support dynamic batching, allowing organizations to maximize GPU efficiency.

API Rate Limits

Pricing is only one consideration.

Organizations should also evaluate:

• Requests per Minute (RPM)
• Tokens per Minute (TPM)
• Concurrent requests
• Burst limits
• Regional quotas

Higher throughput may justify slightly higher API costs for enterprise applications with strict performance requirements.

Example Monthly AI Costs

The following examples illustrate how workload characteristics influence overall AI spending.

Startup AI SaaS

Typical usage:

• Internal chatbot
• Product documentation
• Customer support
• 20–30 employees

Primary cost drivers:

• API usage
• Minimal infrastructure
• Development environment

Managed APIs often provide the lowest total cost at this stage.

AI Coding Assistant

Typical usage:

• Large repositories
• Continuous code generation
• Long prompts
• Multiple developers

Primary cost drivers:

• Input tokens
• Output tokens
• Long context
• Repository analysis

Organizations should evaluate whether API pricing or private deployment offers better long-term value.

Enterprise Knowledge Assistant

Typical usage:

• Internal documentation
• HR policies
• Technical manuals
• Compliance documents

Primary cost drivers:

• Long-context processing
• Retrieval pipelines
• Concurrent users
• GPU infrastructure

Optimizing retrieval quality often has a greater impact on cost than switching between models.

Customer Support Platform

Typical usage:

• Thousands of conversations daily
• Short prompts
• Moderate responses

Primary cost drivers:

• High request volume
• API availability
• Response latency

Here, API stability and throughput are often more important than benchmark differences.

Self-Hosting Costs

Many organizations eventually consider self-hosting to reduce API expenses and improve data privacy.

However, private deployments introduce additional costs.

These include:

• GPU infrastructure
• Storage
• Networking
• Monitoring
• Security
• Maintenance
• Software updates
• Engineering resources

Self-hosting becomes increasingly attractive as AI usage scales, but only if the organization has the expertise to manage production infrastructure.

AWS Deployment Costs

AWS is one of the most popular platforms for enterprise AI deployments.

Typical architecture includes:

Common AWS services include:

• Amazon EKS
• Amazon EC2 GPU instances
• Elastic Load Balancing
• Amazon CloudWatch
• Amazon S3
• IAM
• Amazon ECR

This architecture supports scalable, highly available AI deployments.

GPU Infrastructure Comparison

GPU selection has a major impact on operational costs.

Organizations should select GPUs based on expected throughput rather than choosing the largest available hardware.

Ollama

Ollama is one of the easiest ways to run Qwen and DeepSeek locally.

Ideal for:

• Developers
• Local experimentation
• Offline coding assistants
• Small internal tools

Advantages:

• No API charges
• Simple installation
• Local privacy
• Rapid experimentation

Limitations:

• Limited scalability
• Hardware constraints
• Not intended for large production deployments

vLLM

vLLM has become a leading inference engine for enterprise deployments.

Benefits include:

• High throughput
• Dynamic batching
• Efficient GPU memory usage
• OpenAI-compatible APIs
• Excellent production performance

Organizations running millions of requests often choose vLLM to reduce infrastructure costs while improving response times.

Docker

Docker simplifies deployment by packaging inference servers into portable containers.

Advantages include:

• Consistent environments
• Simplified updates
• Easy testing
• CI/CD integration

Containerized deployments also improve operational reliability across development, staging, and production environments.

Kubernetes

Large-scale deployments typically rely on Kubernetes.

Benefits include:

• Horizontal scaling
• Automatic recovery
• Rolling updates
• Resource scheduling
• High availability
• Multi-region deployment

At EaseCloud, Kubernetes deployments are commonly built on Amazon EKS to support enterprise AI platforms requiring scalability, security, and operational resilience.

Hidden Infrastructure Costs

Many organizations underestimate the ongoing expenses associated with operating AI systems.

Common hidden costs include:

• GPU idle time
• Monitoring and observability
• Security and compliance
• Data transfer
• Storage
• Logging
• Backup systems
• CI/CD pipelines
• Engineering support
• Disaster recovery

These operational expenses can represent a significant portion of the overall AI budget, particularly for self-hosted environments.

EaseCloud Cost Optimization Perspective

When helping organizations deploy Qwen or DeepSeek, EaseCloud focuses on reducing cost per successful inference rather than simply lowering token prices.

Our optimization approach includes:

• Selecting the right model for the workload
• Right-sizing GPU infrastructure
• Implementing intelligent prompt caching
• Optimizing Retrieval-Augmented Generation (RAG)
• Deploying dynamic batching with vLLM
• Autoscaling Kubernetes clusters
• Applying FinOps practices to monitor and control AI spending

By combining infrastructure optimization with workload-specific tuning, organizations can often achieve greater savings than by switching models alone.

API vs Self-Hosting: Which Is More Cost-Effective?

One of the biggest decisions organizations face is whether to use a managed API or deploy AI models on their own infrastructure.

There is no universal answer. The best option depends on workload size, compliance requirements, engineering resources, and long-term growth plans.

The table below summarizes the trade-offs.

When Managed APIs Make Sense

Managed APIs are often the best choice when:

• Building an MVP
• Launching quickly
• Limited engineering resources
• Low or moderate request volume
• Rapid experimentation
• Short development timelines

Benefits include:

• No GPU management
• Automatic updates
• High availability
• Simple integration
• Faster deployment

When Self-Hosting Makes Sense

Self-hosting becomes attractive when organizations require:

• Complete data privacy
• Regulatory compliance
• High request volumes
• Custom fine-tuning
• Internal AI platforms
• Lower long-term inference costs

Self-hosting also enables tighter integration with existing cloud infrastructure and security controls.

EaseCloud Recommendation

At EaseCloud, we often recommend a phased strategy:

Phase 1

Start with managed APIs to validate business value and estimate usage.

Phase 2

As workloads grow, evaluate private deployment on AWS using GPU instances, Amazon EKS, and vLLM.

Phase 3

Optimize infrastructure through autoscaling, intelligent routing, and FinOps practices to reduce long-term operating costs.

This approach minimizes upfront investment while providing a clear path toward scalable enterprise AI.

Cost Optimization Strategies

Reducing AI costs is about improving efficiency rather than simply choosing the lowest-priced model. For a comprehensive framework on reducing AWS cloud costs, see our AWS Cost Optimization Guide.

Below are several proven strategies used in production environments.

Quantization

Quantization reduces model memory requirements without significantly impacting output quality.

Common formats include:

• INT8
• INT4
• FP16
• GGUF
• GPTQ
• AWQ

Benefits include:

• Lower VRAM requirements
• Faster inference
• Reduced GPU costs
• Improved hardware compatibility

For many production workloads, quantized models provide an excellent balance between performance and cost.

Prompt Optimization

Many organizations unknowingly waste tokens.

Common causes include:

• Excessively long system prompts
• Duplicate instructions
• Irrelevant context
• Repeated examples
• Poor Retrieval-Augmented Generation (RAG)

Improving prompt design often reduces monthly AI spending without changing the model.

Best practices:

• Remove unnecessary instructions.
• Keep prompts task-specific.
• Use structured outputs where appropriate.
• Retrieve only relevant documents.

Intelligent RAG

Large context windows do not automatically improve accuracy.

Instead, implement Retrieval-Augmented Generation (RAG) that retrieves only the most relevant information.

Advantages:

• Lower token usage
• Better response quality
• Faster inference
• Reduced API costs

This is one of the highest-impact optimizations for enterprise knowledge assistants. To understand when to use RAG versus fine-tuning, see our guide.

Prompt Caching

Many enterprise applications repeatedly send identical system prompts.

Examples include:

• Company policies
• Coding standards
• AI assistant instructions
• Brand guidelines

Prompt caching reduces repeated processing and can lower costs while improving response times where supported.

Batch Inference

Organizations processing thousands of requests can improve GPU efficiency through batching.

Benefits include:

• Higher GPU utilization
• Lower infrastructure cost
• Increased throughput
• Better scalability

Batching is particularly valuable for offline processing and document analysis workloads.

GPU Sharing

Many GPU deployments operate far below full capacity.

GPU sharing allows multiple applications to use the same hardware, increasing utilization and reducing idle time.

Typical workloads include:

• AI chatbots
• Internal copilots
• Document summarization
• Translation services

Autoscaling

Static GPU clusters often waste money during periods of low demand.

Autoscaling enables infrastructure to expand and contract automatically based on workload.

Benefits:

• Lower operational costs
• Better resource utilization
• Improved availability
• Reduced idle infrastructure

Spot Instances

Cloud providers offer discounted compute capacity through spot instances.

Advantages:

• Significant cost savings
• Suitable for batch inference
• Ideal for model evaluation
• Cost-effective experimentation

However, because spot instances can be interrupted, they are generally best suited for fault-tolerant workloads. For more on using Spot instances, see AWS Fargate Spot Cost Optimization.

Reserved Capacity

Organizations with predictable workloads may benefit from reserved GPU capacity.

Benefits include:

• Lower hourly costs
• Budget predictability
• Long-term savings

Reserved capacity is commonly used for production AI platforms with stable traffic patterns. To understand the trade-offs between AWS purchasing models, see our guide on Reserved vs Spot vs On-Demand.

Multi-Model Routing

Not every request requires the most capable—or most expensive—model.

Multi-model routing directs requests to different models based on complexity.

For example:

• Simple FAQs → Smaller model
• Coding tasks → Qwen Coder
• Advanced reasoning → DeepSeek R1

This strategy improves overall cost efficiency while maintaining response quality.

Which Model Is More Affordable?

The answer depends on your deployment strategy.

Qwen May Offer Better Value When:

• Building multilingual applications
• Creating enterprise knowledge assistants
• Prioritizing documentation quality
• Leveraging Alibaba Cloud services
• Using managed cloud deployments

DeepSeek May Offer Better Value When:

• Building reasoning-intensive applications
• Running competitive programming workloads
• Deploying open-weight models privately
• Optimizing algorithm-heavy AI systems

Neither ecosystem is universally cheaper. Total operating costs depend on API provider, infrastructure choices, optimization techniques, and workload characteristics.

Which Delivers Better ROI?

Return on Investment (ROI) should consider more than infrastructure costs.

A useful evaluation framework includes:

Organizations should evaluate ROI over months or years rather than comparing short-term API expenses.

Recommendations by Organization Type

For Startups

Recommended approach:

• Begin with managed APIs.
• Track monthly token usage.
• Validate product-market fit before investing in infrastructure.
• Optimize prompts before scaling.

For SaaS Companies

Recommended approach:

• Monitor cost per active user.
• Implement caching and batching.
• Consider hybrid deployments as usage grows.

For Large Enterprises

Recommended approach:

• Evaluate Total Cost of Ownership.
• Deploy on Kubernetes.
• Optimize GPU utilization.
• Implement governance and FinOps practices.
• Benchmark models using internal workloads.

Common AI Pricing Mistakes

Many organizations overspend because of avoidable mistakes.

Choosing Based Only on API Prices

Infrastructure, engineering effort, and maintenance often exceed token costs.

Ignoring Prompt Efficiency

Long prompts significantly increase recurring expenses.

Optimize prompts before upgrading infrastructure.

Over-Provisioning GPUs

Buying larger GPUs than necessary increases operating costs without improving application quality.

Deploying Without Monitoring

Without usage analytics, organizations cannot identify inefficient workloads or opportunities for optimization.

Skipping Cost Reviews

AI workloads evolve rapidly.

Regularly review:

• Token usage
• GPU utilization
• Infrastructure costs
• API performance
• Monthly ROI

Conclusion

Pricing is only one part of the AI decision-making process. Organizations that focus exclusively on token costs risk overlooking the larger factors that influence long-term success, including infrastructure efficiency, operational complexity, scalability, and developer productivity.

By evaluating Total Cost of Ownership, optimizing deployment strategies, and aligning infrastructure with business goals, teams can build AI platforms that remain both technically effective and financially sustainable.

Frequently Asked Questions

Is Qwen cheaper than DeepSeek?

It depends on the provider, deployment method, and workload. Managed API pricing and self-hosting costs vary across platforms, so organizations should compare the total operating cost rather than token prices alone.

Which API is more affordable?

Both Qwen and DeepSeek are available through multiple providers with different pricing models. The most affordable option depends on expected token usage, response length, and required service levels.

Can I run both models locally?

Yes.

Both ecosystems support local deployment using tools such as:

• Ollama
• vLLM
• Docker
• Kubernetes
• Hugging Face
• LM Studio

Is self-hosting always cheaper?

Not necessarily.

For low-volume applications, managed APIs are often more economical.

Self-hosting generally becomes attractive as usage increases and organizations have the expertise to operate AI infrastructure efficiently.

Which model is better for startups?

Managed APIs are typically the best starting point, regardless of whether you choose Qwen or DeepSeek. They reduce operational complexity and allow teams to focus on product development.

Which model offers better enterprise value?

Both ecosystems are strong enterprise candidates. The best choice depends on governance requirements, deployment strategy, multilingual needs, reasoning workloads, and existing cloud infrastructure.

Final Verdict

Qwen and DeepSeek both provide excellent value, but the most cost-effective choice depends on how you plan to deploy and scale AI.

Choose Qwen if you prioritize:

• Enterprise software development
• Multilingual applications
• Managed cloud ecosystems
• Documentation generation
• Business automation

Choose DeepSeek if you prioritize:

• Advanced reasoning
• Research-focused workloads
• Algorithm-heavy coding
• Open-weight deployments
• Mathematical problem solving

For many organizations, the winning strategy is not choosing one model over the other but adopting an architecture that allows each model to be used where it delivers the greatest value.