GPT-5.6 (Sol / Terra / Luna) is now evaluated on TrustVector โ€” with day-1 independent verification, incl. METR's benchmark-cheating findings.

Read the evaluation
Evaluation record ยท deepseek-r1

DeepSeek-R1

v20251020

DeepSeek

Modelsupersededcodingreasoningopen-source
84
Strong
About This Model

DeepSeek's standalone reasoning model, now superseded and discontinued. Its reasoning was folded into DeepSeek V3.1's hybrid thinking mode (Aug 2025), then V3.2 (Dec 2025) and V4 (Apr 2026); a successor 'R2' never shipped. The legacy deepseek-reasoner name no longer serves R1 โ€” it routes to V4-Flash's thinking mode and is removed 2026-07-24 (15:59 UTC). R1 remains available only via third-party hosts or self-hosted open weights. Historically 53.6% SWE-bench and 79.8% HumanEval.

Last Evaluated: July 9, 2026
Official Website

Trust Vector Analysis

Dimension Breakdown

๐Ÿš€Performance & Reliability
+

Strong performance with excellent coding capabilities and efficient reasoning. Competitive latency despite reasoning optimization.

task accuracy code

Industry-standard coding benchmarks measuring real-world software engineering tasks

Evidence
SWE-bench Verified โ€” 53.6% resolution rate
HumanEval โ€” 79.8% accuracy on code generation
highVerified: 2026-07-09
task accuracy reasoning

Graduate-level reasoning benchmarks requiring multi-step problem solving

Evidence
MATH Benchmark โ€” 88.5% on mathematical reasoning
GPQA โ€” 58.7% on graduate-level questions
highVerified: 2026-07-09
task accuracy general

Comprehensive knowledge testing across domains

Evidence
MMLU โ€” 74.8% on comprehensive knowledge benchmark
DeepSeek Benchmarks โ€” Strong general performance
highVerified: 2026-07-09
output consistency

Internal testing with repeated prompts at various temperature settings

Evidence
DeepSeek Documentation โ€” Good consistency with reasoning optimization
mediumVerified: 2026-07-09
latency p50

Median latency for API requests with standard prompt sizes

Evidence
DeepSeek Performance Metrics โ€” Typical response time ~1.9s
mediumVerified: 2026-07-09
latency p95

95th percentile response time across diverse workloads

Evidence
Community benchmarking โ€” p95 latency ~3.4s
mediumVerified: 2026-07-09
context window

Official specification from provider

Evidence
DeepSeek Documentation โ€” 64K token context window
highVerified: 2026-07-09
uptime

Historical uptime data from official status page

Evidence
DeepSeek Status โ€” 98.7% uptime (last 90 days)
mediumVerified: 2026-07-09
๐Ÿ›ก๏ธSecurity
+

Good security posture with standard guardrails. Adequate protection for typical use cases.

prompt injection resistance

Testing against OWASP LLM01 prompt injection attacks

Evidence
DeepSeek Safety Documentation โ€” Good resistance to prompt injection
mediumVerified: 2026-07-09
jailbreak resistance

Testing against adversarial prompt datasets

Evidence
DeepSeek Safety Research โ€” Standard safety guardrails
mediumVerified: 2026-07-09
data leakage prevention

Analysis of privacy policies and data handling practices

Evidence
DeepSeek Privacy Policy โ€” Standard data handling practices
mediumVerified: 2026-07-09
output safety

Comprehensive safety testing across harmful content categories

Evidence
DeepSeek Safety Evaluations โ€” Comprehensive safety filtering
mediumVerified: 2026-07-09
api security

Review of API security features and best practices

Evidence
DeepSeek API Documentation โ€” API key authentication, HTTPS, rate limiting
highVerified: 2026-07-09
๐Ÿ”’Privacy & Compliance
+

Moderate privacy posture. Data residency primarily in Asia. Limited compliance certifications for Western markets.

data residency

Review of documentation and privacy policies

Evidence
DeepSeek Documentation โ€” Primary data centers in China and Singapore
mediumVerified: 2026-07-09
training data optout

Analysis of privacy policy and data usage terms

Evidence
DeepSeek Privacy Policy โ€” No training on API data by default
highVerified: 2026-07-09
data retention

Review of terms of service and data retention policies

Evidence
DeepSeek Terms of Service โ€” 60-day default retention
mediumVerified: 2026-07-09
pii handling

Review of data protection capabilities

Evidence
DeepSeek Privacy Documentation โ€” Customer responsible for PII handling
mediumVerified: 2026-07-09
compliance certifications

Verification of compliance certifications

Evidence
DeepSeek Compliance โ€” ISO 27001, limited Western certifications
mediumVerified: 2026-07-09
zero data retention

Review of data handling practices

Evidence
DeepSeek Documentation โ€” No zero retention option
mediumVerified: 2026-07-09
๐Ÿ‘๏ธTrust & Transparency
+

Moderate transparency with standard safety features. Limited disclosure compared to Western providers.

explainability

Evaluation of reasoning transparency

Evidence
DeepSeek Features โ€” Reasoning mode with explanation capabilities
mediumVerified: 2026-07-09
hallucination rate

Testing on factual QA datasets

Evidence
Community Testing โ€” Moderate hallucination rate
mediumVerified: 2026-07-09
bias fairness

Evaluation on bias benchmarks

Evidence
DeepSeek Research โ€” Basic bias mitigation
mediumVerified: 2026-07-09
uncertainty quantification

Assessment of confidence expression

Evidence
Model Behavior โ€” Basic uncertainty expression
mediumVerified: 2026-07-09
model card quality

Review of documentation completeness

Evidence
DeepSeek Documentation โ€” Good technical documentation
highVerified: 2026-07-09
training data transparency

Review of public disclosures

Evidence
DeepSeek Research Papers โ€” Limited disclosure of training data
mediumVerified: 2026-07-09
guardrails

Analysis of built-in safety mechanisms

Evidence
DeepSeek Safety Features โ€” Standard safety guardrails
mediumVerified: 2026-07-09
โš™๏ธOperational Excellence
+

Good operational quality with open licensing. Growing ecosystem with room for maturity.

api design quality

Review of API design and consistency

Evidence
DeepSeek API Documentation โ€” Clean RESTful API design
highVerified: 2026-07-09
sdk quality

Review of SDK quality and maintenance

Evidence
DeepSeek SDKs โ€” Python SDK available, actively maintained
highVerified: 2026-07-09
versioning policy

Review of versioning practices

Evidence
DeepSeek API Documentation โ€” Basic versioning policy
DeepSeek API Pricing and Deprecation Notice โ€” Legacy deepseek-reasoner endpoint deprecates 2026-07-24; R1 line discontinued in favor of V3.1/V3.2/V4 hybrid thinking models
DeepSeek API Change Log โ€” deepseek-reasoner alias currently routes to deepseek-v4-flash thinking mode (no longer serves R1); alias removed 2026-07-24 15:59 UTC, after which requests using it fail
highVerified: 2026-07-09
monitoring observability

Review of monitoring tools

Evidence
DeepSeek Platform โ€” Basic usage dashboard
mediumVerified: 2026-07-09
support quality

Assessment of support options

Evidence
DeepSeek Support โ€” Community support and documentation
mediumVerified: 2026-07-09
ecosystem maturity

Analysis of ecosystem maturity

Evidence
GitHub Community โ€” Growing ecosystem, limited third-party integrations
mediumVerified: 2026-07-09
license terms

Review of licensing terms

Evidence
DeepSeek License โ€” Open license with commercial use allowed
highVerified: 2026-07-09
Strengths
  • +Excellent coding performance (53.6% SWE-bench, 79.8% HumanEval)
  • +Competitive pricing compared to Western alternatives
  • +Good reasoning capabilities with efficient implementation
  • +Open license allowing commercial use
  • +Fast latency (1.9s p50) despite reasoning features
  • +Strong mathematical capabilities
Limitations
  • !Limited data residency options (primarily Asia)
  • !Fewer compliance certifications for Western markets
  • !60-day data retention (not ephemeral)
  • !Limited transparency on training data
  • !Smaller context window (64K tokens)
  • !Less mature ecosystem compared to Western providers
  • !Superseded: line discontinued in favor of DeepSeek V3.1/V3.2/V4 hybrid thinking; legacy deepseek-reasoner alias now routes to V4-Flash (not R1) and is removed 2026-07-24, leaving third-party hosts or self-hosting as the only ways to run R1
Metadata
pricing
input: $0.55 per 1M tokens (historical first-party pricing)
output: $2.19 per 1M tokens (historical first-party pricing)
notes: Historical first-party API pricing no longer applies: the deepseek-reasoner alias now serves V4-Flash at V4 pricing and is removed 2026-07-24. R1 is now served only by third-party hosts (rates vary) or self-hosted from open weights.
last verified: 2026-07-09
context window: 64000
languages
0: English
1: Chinese
2: Japanese
3: Korean
4: Spanish
5: French
6: German
modalities
0: text
api endpoint: https://api.deepseek.com/v1/chat/completions
open source: true
architecture: Transformer-based with efficient reasoning optimization
parameters: Not disclosed

Use Case Ratings

code generation

Excellent coding with 53.6% SWE-bench and 79.8% HumanEval. Strong value proposition with competitive pricing.

customer support

Adequate for customer support but not specialized. Good latency helps.

content creation

Solid content generation capabilities at competitive pricing.

data analysis

Strong analytical capabilities with good reasoning. Excellent value for price.

research assistant

Good research capabilities with reasoning optimization.

legal compliance

Limited compliance certifications for Western markets. Data residency concerns.

healthcare

Not suitable for healthcare due to limited compliance certifications and data residency.

financial analysis

Good analytical capabilities at competitive pricing.

education

Strong tutoring capabilities with good reasoning and affordable pricing.

creative writing

Adequate creative capabilities at good value.