Free tools Windows power users keep installed
One-click scans. No signup required.
Some links on this page are affiliate links: if you buy through them we may earn a commission, at no extra cost to you.
Evaluate an open-weight model as part of a complete business system, not as a download with a benchmark score. Start by defining the task, users, data, jurisdictions, unacceptable failures, response-time and cost limits, and required human controls. Then test candidate weights on representative, held-out examples; inspect provenance and license terms; red-team the real inference stack; measure quality, security, privacy, latency, throughput and operating cost; and establish monitoring, rollback and regulatory controls.
Open weights can permit inspection, fine-tuning and self-hosting, but the label does not prove safety, compliance or production readiness. Microsoft’s 2026 guidance summarizes the governing principle: risk depends on the model, use case, deployment and safeguards. The sections below turn that principle into a repeatable enterprise selection process.
As an Amazon Associate I earn from qualifying purchases.
What is an open-weight AI model, and why does the label matter?
An open-weight model publishes the numerical parameters needed to run the trained network, usually with documentation and a license. That can give an enterprise more control than a hosted black-box API: you may be able to run inference in your own environment, inspect behavior, fine-tune the model, pin a version and choose where data is processed.
Weights are not the same as open source. A release may omit training code, data, preprocessing pipelines, evaluation scripts or complete information about how the weights were produced. License terms can also restrict commercial use, redistribution, modification, fields of use or downstream access. Treat openness as a capability that changes your control options, not as evidence that the system is secure or lawful.
#1 Best Overall
- [Ultra-Lightweight Ear Buds Designed for Small Ears] Each earbud weighs only 3.7g and features a compact, ergonomic in-ear design made especially for small ears. Secure, low-profile, and comfortable for workouts, all-day wear or overnight sleeping.The ultra-slim profile sits flush in your ear with no hard edges poking or rubbing, so you can even sleep on your side without discomfort.
- [Immersive Stereo Sound with TOZO OrigX Technology] TOZO OrigX tuning delivers clear vocals, balanced mids, and natural stereo sound for music, podcasts, and videos.
- [Long Battery Life for Daily Use] Get up to 7 hours of playtime on a single charge, with up to 32 hours total using the charging case—ideal for workdays, commuting, and extended listening sessions.
- [Bluetooth 5.3 & Stable Connection] Bluetooth 5.3 provides fast pairing, stable wireless performance, and reduced dropouts as you move around home or office.
- [Deep Bass with Clear Vocals] High-performance drivers produce punchy bass while keeping vocals clean and detailed for everyday listening.
The deployed product includes much more than the neural network. Permissions, retrieval indexes, system prompts, tools, filters, logging, runtime libraries, model servers, network paths and human-review procedures all affect risk. Microsoft’s Managing open-weight model risk in enterprise deployments (2026) advises assessing the model, application and operating safeguards together.
How do you define the enterprise decision and risk boundary?
Write the decision specification before comparing model cards or leaderboard positions. A precise boundary prevents a high average score from hiding a failure that could injure a customer, expose confidential information or trigger an unauthorized action.
- State the business task. Specify whether the model drafts support replies, extracts fields, summarizes cases, answers questions over a controlled corpus, writes code or calls tools. Define the expected input and output format.
- Name the users and affected people. Include employees, customers, applicants, patients, suppliers and anyone whose data or opportunities may be affected, not just the person operating the interface.
- Classify the data. Mark public, internal, confidential, personal, regulated and security-sensitive data. Record whether prompts, retrieved documents, tool arguments and logs may contain each class.
- Set geography and jurisdiction. Record where users, data, inference hardware and support personnel are located. Residency, transfer and sector rules can change the acceptable architecture.
- Define unacceptable failures. Examples include fabricated eligibility decisions, disclosure of another customer’s record, unsafe instructions, unauthorized transactions, discriminatory recommendations or a tool call made without confirmation.
- Set service limits. Establish maximum latency, minimum throughput, availability target, context length, peak-load behavior and cost per completed task. These are acceptance criteria, not observations to collect after launch.
- Specify escalation. Decide when the system must refuse, ask for clarification, route to a trained reviewer or disable a tool. Assign a named owner for each escalation path.
Keep two artifacts separate: a model evaluation record and an application risk assessment. A model can perform well in isolation while the surrounding retrieval, permissions or automation makes the complete application unsafe.
What provenance and legal checks should you complete?
Before running confidential prompts, create a provenance record for every candidate and every later fine-tune. The record should be detailed enough for procurement, security, legal and an incident investigator to reconstruct what was used.
- Publisher and release identity: capture the organization, exact model name, parameter variant, quantization, release date and immutable version or commit.
- Model and training disclosures: save the model card, architecture description, known limitations, intended uses, disallowed uses, evaluation methods and any available training-data summary.
- License obligations: verify commercial-use rights, redistribution and hosting rights, modification permissions, attribution, notice requirements, acceptable-use restrictions and obligations passed to downstream users.
- Data rights: determine whether your fine-tuning or retrieval data may be used, retained, transferred or inspected by any service involved in preparation and inference.
- Security provenance: record artifact hashes, download location, dependency versions and the process used to scan packages and weights for tampering.
- Support and change policy: note how vulnerabilities, model revisions and license changes are announced, and who is responsible for acting on them.
Meta recommends using model cards and creating one for a custom model. Your internal card should add the business purpose, data classes, evaluation results, residual risks, operating limits, approved users and rollback version. If a disclosure is missing, mark it as unknown and treat that uncertainty as a governance risk rather than filling it with assumptions.
How do you build a representative enterprise test set?
Public benchmarks are useful starting points, but production decisions require held-out examples that resemble your own workflow. Build the set from historical tasks only after removing personal information or obtaining the required permissions, and keep a sealed portion that neither prompt authors nor model tuners can inspect.
Include normal work and difficult edges
- Typical requests across each important department, product line and language.
- Long documents, tables, mixed formats and retrieval questions that require citing the correct passage.
- Ambiguous, incomplete or contradictory instructions that should trigger clarification.
- Tool-use sequences with valid, invalid and intentionally dangerous arguments.
- Privacy probes that attempt to recover secrets from prompts, context, memory or logs.
- Prompt-injection and jailbreak attempts delivered directly and through retrieved documents.
- Fairness slices for relevant demographic, geographic, linguistic or accessibility groups.
- Refusal cases where a safe response is required, plus near-boundary cases where refusal would block legitimate work.
- Operational cases such as peak concurrency, context-window limits, retries and partial tool failures.
Label each example with the expected answer or acceptable range, evidence that must be cited, prohibited content, severity of an error and whether a human must review it. Separate tuning data from evaluation data. Reusing the same examples for prompt optimization and final scoring produces an optimistic result.
Rank #2
- 【IP68 Waterproof Multifunctinal Smart Watch】Torjalph fitness activity tracker smart watch is IP68 waterproof technology ,you can wear it when swimming,handwashing,it's rain-proof which means that you perform multiple exercise modes outside without worrying about sweat or rain corroding your fitness smart watch,and this watch equipped with a 230 mAh large capacity battery magnetic charging ,and this sports activity tracker watch can be used 5-7 days If fully charged.
- 【Larger 1.83'' Full Screen Watch & Multi-dial Selection】1.83 inch full HD color touch screen smart watch for men / women provides excellent 240*284 resolution image quality ,This Bluetooth smart watch offers a better touch experience and remarkable immersion and ,more than 200 watch faces for you to choose from App"Da Fit ", you can change the dial style according to your mood to wear keeping the world on your wrist
- 【Accurate Activity Tracker Smart Watch 】sports watch activity tracker uses sweat and dust resistant breathable silicone strap, our Fitness Tracker Smart Watch for Android iOS Phones supports 12 sports modes,walking,running,bicycle,swimming,badminton,basketball,football,rope skipping. This bluetooth calls smart watch for men women is lightweight and stylish, bringing more convenient and relaxing sports experience while also providing more protection.
- 【Your 24 Hour Personal Daily Health Monitor】Fitness watch for women men full touch screen IP67 waterproof bluetooth smartwatch for Android iOS Phones with heart rate blood pressure monitor multifunction Andriod smart watch with steps counter sleep monitor sports fitness watch, monitoring blood oxygen saturation allows you to understand your physical condition in real time to track fitness progress ,to know your health status in time.
- 【More Functions Bluetooth Call Smart Watch】This fitness activity smart watch full screen for iPhone and Android phones includes a bunch of new features including:The bluetooth smart watch can hang up the phone, receive and display WhatsAPP /Text /Facebook /Twitter /Messenger reminder.Support sedentary reminder, alarm clock, stopwatch, timer,weather information,music control, remote camera control compatible with Android 5.0 and above or iOS 9.0 and above smartphones.
Which benchmarks matter for enterprise LLM selection?
Use public evaluations to discover strengths and weaknesses, not to issue a universal safety certificate. Meta identifies Stanford HELM, EleutherAI’s evaluation harness, CyberSecEval and Hugging Face datasets as useful starting resources; validate every important finding on your own held-out data.
| Evaluation source or method | What it can tell you | What it cannot establish |
|---|---|---|
| Stanford HELM | Comparable results across documented scenarios and metrics. | Performance on your private workflow, prompts, retrieval corpus or tool permissions. |
| EleutherAI evaluation harness | Repeatable execution of many academic and task-specific benchmarks. | That a benchmark is aligned with your business harm or that scores remain stable after quantization. |
| CyberSecEval | Security-oriented indicators for selected cyber-use and misuse scenarios. | Complete resistance to your attack surface, plugins, agents or confidential data. |
| Hugging Face datasets and other public sets | Candidate prompts, labels and domain slices for test construction. | Legal permission to use every example or representativeness for your users. |
| Internal held-out suite | Task quality, groundedness and failure rates under your exact instructions and data. | Protection from future model, prompt, infrastructure or user-behavior changes unless continuously rerun. |
Record the dataset version, prompt template, decoding settings, hardware, software stack and date for every run. Without those details, two reported scores may not be comparable.
How should you measure quality, safety and operating performance?
Use a metric set tied to the risks in your decision specification. A single accuracy number can reward fluent hallucinations, conceal subgroup failures and ignore the cost of serving the model.
| Dimension | Measurements to collect | Useful interpretation |
|---|---|---|
| Task quality | Exact-match or structured-field accuracy, rubric scores, ranking quality, edit distance and human preference where appropriate. | Whether the output completes the defined job, not whether it merely sounds plausible. |
| Factuality and groundedness | Claim accuracy, citation precision, retrieval recall, unsupported-claim rate and abstention quality. | Whether answers are supported by approved evidence and whether uncertainty is handled honestly. |
| Safety and misuse | Harmful-output rate, refusal precision and recall, jailbreak success, prompt-injection success and unsafe tool-call rate. | How often the system fails in the ways your risk assessment considers severe. |
| Privacy | Canary-secret extraction, memorization probes, cross-tenant leakage tests, retention behavior and log exposure. | Whether confidential information can cross the boundaries you defined. |
| Fairness | Quality and refusal differences across relevant groups, languages and locations. | Whether a single aggregate score hides material disparities. |
| Latency and capacity | Time to first token, total response time, tokens per second, queue time, concurrency and tail percentiles. | Whether user experience and service-level objectives hold under realistic load. |
| Resource efficiency | Memory footprint, accelerator and CPU utilization, power where measured, context-window behavior and cost per completed task. | Whether self-hosting economics remain acceptable at expected and peak volume. |
| Reliability | Error rates, timeout frequency, restart behavior, uptime and recovery time. | Whether the service can be operated predictably rather than only demoed successfully. |
Define thresholds by severity. For example, a minor formatting defect may be tolerated, while any unauthorized disclosure or irreversible tool action may be a zero-tolerance gate. Microsoft recommends connecting each evaluation to an identified risk and mitigation instead of treating a benchmark as a general safety score.
How do you test the model in its real deployment stack?
After isolated model tests, exercise the complete application in a controlled environment that mirrors production. The same weights can behave differently when a system prompt, retrieval pipeline, quantizer, model server, filter or tool wrapper changes.
- Reproduce the production path. Use the intended serving runtime, quantization, context assembly, retrieval index, authentication, rate limits, filters and logging configuration.
- Test permissions. Attempt access with every user role, expired credential, tenant boundary and service-account scope. Confirm that the model cannot grant itself authority through text.
- Attack retrieval and tools. Place malicious instructions in documents, manipulate metadata, send malformed tool results and test confirmation requirements for high-impact actions.
- Probe data handling. Verify encryption, retention, redaction, log access, backup behavior and deletion workflows for prompts, outputs, embeddings and traces.
- Run adversarial and abuse exercises. Combine jailbreaks, prompt injection, multilingual requests, long contexts and repeated attempts. Record successful attack paths and time to detection.
- Validate human review. Measure whether reviewers receive enough context, can override the model, see the evidence used and stop an in-progress action.
- Document limitations. Publish known failure modes, unsupported languages, context limits, tool restrictions and conditions under which the service must not be used.
The UK National Cyber Security Centre’s 2023 secure-deployment guidance calls for appropriate security evaluation, including benchmarking and red teaming, before release. It also emphasizes clear limitations and explicit responsibility for data use and security; assign those responsibilities in writing rather than assuming the model publisher owns them.
What should you compare when two candidates remain?
Use a weighted decision matrix only after hard safety, legal and operational gates have been passed. Weights should reflect the business harm of failure, not the convenience of measuring a metric.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
Rank #3
- [Low Noise High Fidelity] New bee guitar cord with 22AWG oxygen-free-copper center conductors provides enhanced signal transfer, the PVC insulation material and the high-density stranded copper double shield provide noise rejection for quiet operation
- [Ideal for Choice] New bee 1/4 straight and right angle metal connectors, ensures a secure connection in any instrument, pedal, or amp jack. L shape great fro on stage guitar stands
- [Wide Compatible] Suitable for mixer, power amplifier, microphone, effector, decoder, equalizer, electoronic organ, electric guitar, electric bass and other electronic equipments. Widely used in the installation of smart home, karaoke speaker, professional stage audio, etc
- [Easy Disassemble] Making incidental repairs quick and easy! Very nice, flexible cable
- [Dont Hesite to Order] 24 hours customer services and professional technology team are standing by
| Comparison axis | Questions to ask |
|---|---|
| Task quality and groundedness | Does it solve the priority workflow, cite approved evidence and abstain appropriately? |
| Safety and abuse resistance | How does it perform against your harmful-output, jailbreak, injection and unsafe-action tests? |
| Privacy and data handling | Can you enforce isolation, retention, deletion, encryption and least-privilege access? |
| License and provenance | Are commercial use, modification, redistribution and downstream obligations acceptable and documented? |
| Documentation quality | Are limitations, training disclosures, evaluation methods and version history sufficient for audit? |
| Latency and throughput | Does the chosen hardware meet tail-latency and peak-concurrency requirements? |
| Hardware and operating cost | What is the measured cost per successful task, including idle capacity, storage, monitoring and support? |
| Fine-tuning and customization | Can you adapt behavior without losing safety, provenance or rollback capability? |
| Ecosystem and support | Are serving tools, patches, evaluation integrations and skilled operators available? |
| Deployment control | Can you pin versions, choose locations, restrict network access and reproduce builds? |
| Monitoring and incident response | Can you detect abuse, investigate a trace, disable a tool and restore a known-good version? |
| Regulatory fit | Can the provider and your organization supply the documentation, oversight and reporting required in each jurisdiction? |
Publish the matrix with raw measurements, test-set versions, confidence intervals or uncertainty notes where available, and an explanation for every weight. A model with a lower average score may still be the better choice if it clears high-severity safety gates and is easier to control.
What must you plan for self-hosting and production operations?
Self-hosting moves responsibilities to your organization; it does not remove them. Estimate capacity from measured memory use, context length, concurrency, batching and redundancy on the exact serving stack you intend to operate. Include hardware acquisition or rental, storage, networking, power, observability, patching, on-call coverage and disaster recovery in the cost-per-task model.
- Pin the model artifact, tokenizer, runtime, drivers and container images.
- Separate development, evaluation, staging and production environments.
- Restrict outbound network access and model-server privileges to the minimum required.
- Encrypt data in transit and at rest, and segregate tenants and environments.
- Capture immutable audit events for access, prompt policy decisions, tool calls, overrides, model version and configuration.
- Monitor quality proxies, refusal rates, latency, queue depth, resource saturation, unusual usage and security alerts.
- Maintain a tested rollback image and a procedure for disabling retrieval or tools independently of the model.
Fine-tuning introduces another change surface. Keep the base-model identity, training data lineage, hyperparameters, evaluation results and approval record for each adapter or merged checkpoint. Do not promote a tuned model merely because its in-sample score improved.
How should governance and change control work?
Document ownership before launch. A practical operating agreement names the model owner, application owner, data owner, security lead, legal or compliance contact and incident commander, with authority to pause service.
Recommended Free Tools
- Version control: assign immutable identifiers to weights, prompts, retrieval indexes, tool schemas, filters and infrastructure.
- Pre-release review: require the critical evaluation suite, security tests, license review and human-oversight sign-off for every material change.
- Continuous monitoring: sample outputs under approved privacy rules, track drift in task mix and failure rates, and alert on new attack patterns.
- Incident response: define severity levels, evidence preservation, customer notification criteria, containment steps and regulator contacts where required.
- Rollback and retirement: rehearse restoration of the prior version, revoke compromised credentials, preserve audit records and securely delete retired artifacts when policy requires.
Re-run critical evaluations after changing the model, prompt, retrieval corpus, tool, filter, quantization, serving runtime, hardware or user-facing workflow. A silent infrastructure change can alter behavior even when the weights are identical.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.What EU regulatory scope should an enterprise check?
EU obligations depend on the role of the organization, the system’s use and the model’s classification. Do not assume that a self-hosted model is outside the AI Act.
- GPAI provider duties: European Commission guidance states that provider obligations began applying on 2 August 2025. Providers must maintain technical documentation and give downstream users information needed to understand and integrate the model. Copyright policy and a summary of training content are also part of the Commission’s GPAI guidance.
- Systemic-risk GPAI: Models meeting the Commission’s criteria, including the guidance FAQ’s compute threshold of 1023 floating-point operations for scope clarification, face additional evaluation, incident-reporting and cybersecurity duties.
- Downstream systems: An enterprise integrating a model must assess whether its application is a regulated high-risk system or falls under another category. The Commission’s AI Act overview lists risk management, data quality, logging, documentation, human oversight, robustness, cybersecurity and accuracy among high-risk requirements.
- Timeline: The Commission lists 2 August 2026 for key enforcement powers and 2 December 2027 for the listed high-risk AI rules under the current timeline. Confirm the latest implementation details with qualified counsel because delegated acts and national procedures can affect practical duties.
Keep evidence that connects each control to the applicable requirement: risk assessments, data governance records, evaluation reports, logs, human-oversight procedures, cybersecurity tests, incident records and technical documentation. Geographic deployment and the organization’s role can change the analysis.
Rank #4
- Digital Family Calendar: BSIMB digital calendar wall touch screen puts an end to your chaotic life. Designed for busy families, the digital calendar keeps you organised and in step with your family. Assign different colours and tasks to each family member to make family scheduling easier and more intuitive.
- Easy Setup and Auto-Sync: Plug in, connect to WLAN and auto-syn your electric calendar in minutes. Digital wall calendar compatible with Google calendar, iCloud calendar, Outlook calendar, Cozi calendar and Yahoo calendar. Add events directly on the device or through the ecalendar app. Electronic calendar stays current whether you're at work or traveling—no manual updates needed.
- All-In-One Home Assistant: Smart calendar builds habits and encourage independence with personalised chores. Display daily meal plan to reduce the daily ‘What's for dinner?’ stress. To-do lists keep track of everything you want. Turning chores into valuable achievements with stars and rewards, making life easier and more enjoyable.
- Manage Anytime Anywhere: BSIMB electronic calendar for family stay connected with the powerful ecalendar app. Access and manage your digital wall calendar anytime, anywhere with free features like stars & rewards, photo screensaver, chore charts, meals plans, to-do lists etc. Electronic planner paid features like Magic Import tailor advanced tools for users with special needs.
- Stay in Touch with Loved Ones: Share photos or videos from your phone to electronic wall calendar with app, instantly update your life moment, and keep in touch with your family, especially those who are far away; you can also surprise your smart calendar digital wall by sending him a wish card; it's also great for keeping tech-illiterate elders in touch with the youngsters.
What decision framework produces a defensible go or no-go?
Use a staged gate rather than a single weighted score. The sequence prevents an attractive quality result from compensating for a disqualifying legal or safety defect.
The Tool Desk
Outbyte Driver Updater FREEFix the driver behind crashes, sound loss and screen glitchesFind Drivers →Outbyte PC Repair FREEClear out junk files and repair common Windows errorsFree Scan →- Gate 1 — Scope: the task, users, data, harms, jurisdictions, service limits and escalation path are documented.
- Gate 2 — Provenance: publisher disclosures, license rights, artifact identity, training-data information and downstream obligations are acceptable or explicitly escalated.
- Gate 3 — Evidence: the candidate completes representative held-out tests, including edge, multilingual, retrieval, tool, privacy and adversarial cases.
- Gate 4 — Safety: high-severity failure thresholds are met, with no unresolved path to unauthorized disclosure or irreversible action.
- Gate 5 — Operations: latency, throughput, memory, reliability, monitoring, support and cost meet the defined limits on the intended stack.
- Gate 6 — Governance: owners, human review, auditability, incident response, rollback and re-evaluation triggers are funded and rehearsed.
- Gate 7 — Regulatory: the organization can satisfy applicable documentation, oversight, security and reporting duties before the planned launch date.
Record three outcomes for each candidate: approved for the specified use, approved with controls and restricted scope, or rejected with the failure evidence. An approval should name the exact model version and deployment configuration; it should not silently cover future revisions.
What common evaluation mistakes should you avoid?
Choosing by leaderboard rank
Public scores may use different prompts, data, decoding settings and contamination controls. Use them to shortlist candidates, then rely on your sealed enterprise suite and reproducible settings.
Confusing fluent text with grounded answers
Require evidence citations, claim-level checks and calibrated abstention. A polished unsupported answer is a factuality failure even when users prefer its style.
Testing the model but not the application
Red-team retrieval, permissions, tools, logs and human hand-offs together. Most serious failures occur at interfaces between components.
Ignoring license and disclosure gaps
Unknown training provenance, ambiguous redistribution language or missing limitation details should trigger legal and procurement review, not an assumption that the most permissive interpretation is safe.
Optimizing an average while missing a severe slice
Report subgroup, language, refusal, privacy and high-impact action results separately. Set hard gates for failures that cannot be traded away for gains elsewhere.
Failing to retest after changes
Quantization, a new retrieval index, a system-prompt edit or a runtime upgrade can change outputs and attack resistance. Treat each as a release event when it can alter risk.
What is the practical bottom line?
The strongest enterprise candidate is not necessarily the largest or highest-scoring model. It is the model and deployment combination that clears your defined safety and legal gates, performs reliably on representative work, fits latency and cost limits, and can be monitored, explained, rolled back and governed over time. Open weights are valuable when that control reduces a real business or regulatory risk; they are not a substitute for evidence, security engineering or accountable operations.
Quick wins for a faster PC:
Repair Windows errors before they cause bigger problemsFix Now →Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →Frequently Asked Questions
Can a smaller open-weight model be better for an enterprise than a larger one?
Yes, if it meets the required task quality and safety thresholds with lower latency, memory use or operating cost. The decision should come from the same held-out tests and deployment measurements, not parameter count alone.
How often should an enterprise rerun model evaluations?
Rerun critical tests for every model, prompt, retrieval, tool, filter, quantization, runtime or infrastructure change that could alter behavior. Also schedule periodic reviews to catch drift in user requests, attack techniques and regulatory requirements.
What evidence should procurement request from an open-weight model publisher?
Request the exact release identity, model card, license, intended and prohibited uses, known limitations, evaluation methodology, training-data disclosures, security-contact process and change-notification policy. Preserve the documents with the approved version so later audits can distinguish what was known at purchase time.
Is self-hosting enough to keep enterprise data private?
No. Self-hosting can reduce external data transfer, but privacy still depends on tenant isolation, permissions, retention, logs, backups, retrieval indexes, tool calls and operator access. Test those boundaries in the complete application and verify deletion and incident-response procedures.
Crashes, No Sound, or Screen Glitches?
Random freezes, missing sound and display glitches usually trace back to one bad driver. Find and replace yours safely.Free scan · under a minuteWindows Errors? Fix Them Before They Spread
Repair common Windows errors and clear accumulated junk for a smoother, more stable PC - no reinstall needed.Free scan · no reinstallQuick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
