Content libraryOperating

Vendor Evaluation Scorecard

Scorecard for evaluating LLM API vendors across price, quality, security, and ops.

MarkdownOperating

LLM API Vendor Evaluation Scorecard

Version: 1.0
Evaluation Date: ____________________
Evaluator(s): ____________________
Review Cycle: ☐ New Vendor Selection ☐ Annual Re-evaluation ☐ Contract Renewal


How to Use This Scorecard

  1. Complete one scorecard per vendor being evaluated
  2. Use consistent data sources across all vendors (same benchmark tasks, same time period)
  3. Weight criteria based on your organization's priorities before scoring
  4. Score each criterion 1–5 using the rubric provided
  5. Calculate weighted scores and compare vendors side by side
  6. Attach supporting evidence (benchmark results, pricing quotes, compliance certs) as appendices

Scoring Rubric:

Score Label Definition
1 Unacceptable Does not meet minimum requirements
2 Below Average Meets some requirements with significant gaps
3 Acceptable Meets core requirements adequately
4 Strong Exceeds requirements in meaningful ways
5 Exceptional Best-in-class; clear competitive advantage

1. Provider Overview

Field Details
Provider Name
Website
Headquarters / Jurisdiction
Year Founded
Funding / Public Status
Enterprise Customers (reference-able)
Primary Contact (Sales)
Primary Contact (Technical)
Evaluation Period Start: ________ End: ________
Models Under Evaluation

Provider Maturity Signals

  • SOC 2 Type II certified
  • Listed on major cloud marketplaces (AWS, Azure, GCP)
  • Published uptime history / status page
  • Public model cards / technical documentation
  • Enterprise reference customers in our industry
  • Dedicated enterprise support tier available
  • Published security whitepaper

2. Pricing Comparison

2.1 Per-Token Pricing

Model Input Rate ($/1M tokens) Output Rate ($/1M tokens) Output Premium Context Window Max Output Length

2.2 Volume Tier Discounts

Monthly Spend Tier Discount Effective Input Rate Effective Output Rate Commitment Required
$0 – $10K 0% None
$10K – $50K __%
$50K – $200K __%
$200K – $1M __%
$1M+ __%

2.3 Batch / Async Pricing

Feature Available? Discount vs. Real-Time SLA (turnaround) Max Batch Size
Batch API ☐ Yes ☐ No __%
Cached Input ☐ Yes ☐ No __% N/A N/A
Prompt Caching ☐ Yes ☐ No __% N/A

2.4 Free Tier / Credits

Item Details
Free tier available? ☐ Yes ☐ No
Free tier monthly token allowance
Startup / enterprise credits
Credit expiration
Restrictions on free-tier usage

2.5 Blended Cost Projection

Use your organization's actual or projected workload mix:

Workload Monthly Volume (M tokens) Input % Output % Model Monthly Cost
Customer Support Bot 75% 25% $
Document Extraction 85% 15% $
Code Generation 60% 40% $
Batch Enrichment 90% 10% $
Total $

3. Performance Metrics

3.1 Latency

Measured under your representative workload (specify prompt length and output length):

Metric Target Measured Value Score (1–5)
Time to First Token (p50) < 200ms
Time to First Token (p95) < 500ms
Time to First Token (p99) < 1,000ms
End-to-End Latency (p50) < 1.5s
End-to-End Latency (p95) < 3s
End-to-End Latency (p99) < 5s
Tokens per Second (generation speed) > 80 tok/s

3.2 Throughput & Capacity

Metric Details
Requests per Minute (RPM) limit
Tokens per Minute (TPM) limit
Tokens per Day (TPD) limit
Concurrent request limit
Burst capacity above limits
Rate limit increase process
Time to provision higher limits

3.3 Reliability & Uptime

Metric Target Measured / Published Score (1–5)
Monthly uptime SLA 99.9%
Actual uptime (last 6 months) > 99.9%
Mean Time to Recovery (MTTR) < 30 min
Incident frequency (last 6 months) < 3 incidents
Degraded performance events < 5 events
Status page transparency Public, real-time

4. Quality Assessment

4.1 Accuracy by Task Type

Benchmark each model on your actual use cases. Use ≥ 100 test cases per task.

Task Type Test Set Size Metric Target Model A Score Model B Score Notes
Classification 200 Accuracy (%) > 95%
Entity Extraction 200 F1 Score > 0.90
Summarization 100 ROUGE-L / Human Eval > 0.85
Code Generation 100 Pass@1 (%) > 80%
Complex Reasoning 100 Human Eval (1–5) > 4.0
RAG + Q&A 150 Answer Accuracy (%) > 90%
JSON Compliance 200 Valid JSON (%) > 99%

4.2 Output Quality Characteristics

Characteristic Rating (1–5) Evidence
Instruction following
Hallucination rate (lower = better)
Format compliance / structured output
Consistency across identical prompts
Multilingual capability
Reasoning chain quality
Refusal rate on valid requests

5. Security & Compliance

5.1 Certifications & Standards

Requirement Status Expiry / Audit Date Evidence
SOC 2 Type II ☐ Yes ☐ No ☐ In Progress
SOC 3 ☐ Yes ☐ No
ISO 27001 ☐ Yes ☐ No ☐ In Progress
ISO 27701 (Privacy) ☐ Yes ☐ No
HIPAA BAA available ☐ Yes ☐ No ☐ N/A
FedRAMP ☐ Yes ☐ No ☐ In Progress
PCI DSS ☐ Yes ☐ No ☐ N/A

5.2 Data Privacy & Residency

Requirement Status Details
GDPR compliant ☐ Yes ☐ No
CCPA compliant ☐ Yes ☐ No
Data residency options (US, EU, APAC)
Data processing agreement (DPA) available ☐ Yes ☐ No
Customer data used for training ☐ Yes ☐ No ☐ Opt-out
Data retention policy
Right to deletion / purge ☐ Yes ☐ No
PII detection and redaction ☐ Yes ☐ No

5.3 Network & Encryption

Feature Status Details
TLS 1.3 in transit ☐ Yes ☐ No
Encryption at rest (AES-256) ☐ Yes ☐ No
Customer-managed encryption keys (CMEK) ☐ Yes ☐ No
VPC / Private Link connectivity ☐ Yes ☐ No
IP allowlisting ☐ Yes ☐ No
API key rotation support ☐ Yes ☐ No
SSO / SAML for management console ☐ Yes ☐ No
Audit logging (API access logs) ☐ Yes ☐ No

6. Operational Features

6.1 API & Developer Experience

Feature Status Score (1–5)
REST API ☐ Yes ☐ No
Streaming support (SSE) ☐ Yes ☐ No
Function / Tool calling ☐ Yes ☐ No
Structured output (JSON mode) ☐ Yes ☐ No
Vision / multimodal support ☐ Yes ☐ No
Embeddings API ☐ Yes ☐ No
Fine-tuning support ☐ Yes ☐ No
Model versioning / pinning ☐ Yes ☐ No
Deprecation notice period ____________
SDK quality (Python, JS, Go, etc.) ☐ Official ☐ Community
OpenAI-compatible API surface ☐ Yes ☐ Partial ☐ No

6.2 Monitoring & Observability

Feature Status Score (1–5)
Usage dashboard ☐ Yes ☐ No
Real-time token consumption API ☐ Yes ☐ No
Cost breakdown by API key / project ☐ Yes ☐ No
Custom metadata / tags on requests ☐ Yes ☐ No
Webhook / alert integration ☐ Yes ☐ No
Spending alerts and limits ☐ Yes ☐ No
Exportable usage reports (CSV/JSON) ☐ Yes ☐ No

6.3 Support

Feature Details Score (1–5)
Support tiers available
Enterprise support SLA (response time) P1: ____ P2: ____ P3: ____
Dedicated account manager ☐ Yes ☐ No (min spend: $____)
Technical onboarding assistance ☐ Yes ☐ No
Community / forum support ☐ Active ☐ Limited ☐ None
Documentation quality

7. Contract Terms

Term Details Assessment
Minimum commitment period
Annual spend commitment
Prepaid credits required ☐ Yes ☐ No
Overage pricing (above commitment)
Price protection / rate lock
Termination for convenience ☐ Yes ☐ No — notice period: ____
Data portability on exit
Model deprecation guarantees
SLA credit mechanism
Auto-renewal terms
Liability cap
Indemnification (IP / output) ☐ Yes ☐ No
Insurance (E&O, Cyber) ☐ Yes ☐ No

Contract Risk Assessment

Risk Severity Mitigation
Vendor lock-in (proprietary API) ☐ High ☐ Med ☐ Low
Price increase exposure ☐ High ☐ Med ☐ Low
Model discontinuation risk ☐ High ☐ Med ☐ Low
Data sovereignty risk ☐ High ☐ Med ☐ Low
Single-vendor dependency ☐ High ☐ Med ☐ Low

8. Weighted Scoring Matrix

Adjust weights to match your organization's priorities (weights must sum to 100).

# Criteria Weight (%) Score (1–5) Weighted Score
1 Pricing & Cost Efficiency 20
Per-token pricing competitiveness
Volume discount depth
Batch / caching discounts
2 Performance 20
Latency (TTFT + E2E)
Throughput / rate limits
Uptime / reliability
3 Quality 20
Accuracy on target tasks
Structured output compliance
Hallucination rate
4 Security & Compliance 15
Certifications (SOC2, ISO, etc.)
Data privacy & residency
Encryption & network security
5 Operational Features 10
API capabilities & DX
Monitoring & observability
Support quality
6 Contract & Commercial 10
Flexibility (commitment, exit)
Price protection
IP indemnification
7 Strategic Fit 5
Roadmap alignment
Multi-model strategy support
Ecosystem / marketplace integration
TOTAL 100 /5.00

9. Comparative Summary

Complete after evaluating all vendors:

Dimension Vendor A Vendor B Vendor C
Weighted Total Score /5.00 /5.00 /5.00
Projected Monthly Cost $ $ $
Projected Annual Cost $ $ $
Top Strength
Top Risk
Recommendation ☐ Primary ☐ Secondary ☐ Reject ☐ Primary ☐ Secondary ☐ Reject ☐ Primary ☐ Secondary ☐ Reject

10. Evaluation Sign-Off

Role Name Signature Date
Engineering Lead
FinOps / Finance
Security / Compliance
Product Owner
VP / Director (Approver)

Decision Record

Selected Vendor(s): ____________________
Rationale: ____________________
Contract Start Date: ____________________
Next Re-evaluation Date: ____________________

Appendices Checklist

  • A: Benchmark test results (raw data)
  • B: Pricing quotes / proposals from each vendor
  • C: Security questionnaire responses
  • D: Reference customer call notes
  • E: Contract redline / legal review notes

Template from the TokenOps Atlas — tokenops-atlas