Quick answer. DeepSeek V3 has a weighted score of 80 and cost efficiency of 98 at $0.27 in and $1.10 out per million tokens. Privacy is 52. GDPR, HIPAA and SOC 2 are false on the hosted record. It is self-hostable. Choose the hosted API only when cost dominates and the data class tolerates China jurisdiction; otherwise self-host or pick another row.
DeepSeek V3 is the open-weight budget row in the published Best AI Match dataset, ranked 16 of 21 with a weighted score of 80. This profile decides whether the lowest Wave 2 API rates justify the privacy and jurisdiction trade. It is not a recommendation to send customer data to any China-jurisdiction hosted API by default. Cost efficiency is 98. Task performance is 85. Privacy is 52 and Truth Score is 68. Safety is 66. Context is a 128,000-token window, scored 78. The model is self-hostable, which is the control that can change the privacy story if you operate the weights yourself. Jurisdiction on the published record is China. API rates are $0.27 per million input tokens and $1.10 per million output tokens. GDPR, HIPAA and SOC 2 are all marked false for the hosted path. Perspective Score is minus 6 and is excluded from the weighted score. Best-fit jobs in the dataset are cost-optimised coding, budget API workloads, self-hosted deployment and open-weight research. Read those jobs against the compliance flags before you treat the sticker price as the decision.
Budget price is not a privacy waiver
Cost efficiency of 98 at $0.27 and $1.10 per million tokens is why this row appears on cheap-API shortlists. Privacy 52 and false GDPR, HIPAA and SOC 2 flags are why it fails many customer-data briefs on the hosted path. The head-to-head with OpenAI workhorses is on ChatGPT vs DeepSeek. Risk framing for Chinese models is on Chinese AI risk assessment and DeepSeek explained.
| Cost versus risk lens | Published figure |
|---|---|
| Weighted score (rank 16 of 21) | 80 |
| Cost efficiency | 98 |
| API price, input / output per million | $0.27 / $1.10 |
| Task performance | 85 |
| Privacy | 52 |
| Truth Score | 68 |
| Safety | 66 |
| Self-hostable | true |
| Jurisdiction | China |
| GDPR / HIPAA / SOC 2 | false / false / false |
Figures from /data/ai-comparison-2026.json. Last verified 2026-06-26. Scoring version v1.0. Affiliate links are off.
When self-hosting changes the story
Self-hostable true means you can move the weights onto infrastructure you control. That can raise the privacy posture relative to the hosted API score of 52, but only if operations, access and logging match the data class. The decision frame is on self-hosted vs API AI and when a cheaper model is safer for customer data.
Truth and coding jobs
Truth Score is 68 and task is 85. Cost-optimised coding is on the best-fit list, which is why teams sometimes accept the hosted risk for non-sensitive code. Customer data and regulated text are a different data class. See best AI for coding and the privacy checklist on AI data privacy.
Official page, affiliates off
Perspective Score is −6. It does not move the weighted score. The official site is deepseek.com. That link is not an affiliate link.
Match with the privacy constraint on. Use the match engine when cost and privacy both matter. Compare Mistral Large 3 for an EU self-host path, or return to the model index.
FAQ
What is DeepSeek V3's weighted score?
DeepSeek V3 has a weighted score of 80 and is ranked 16 in the published Best AI Match dataset, last verified 2026-06-26.
How much does the DeepSeek V3 API cost?
The dataset lists $0.27 per million input tokens and $1.10 per million output tokens.
Can DeepSeek V3 be self-hosted?
Yes. The published record marks DeepSeek V3 as self-hostable. The hosted API still shows China jurisdiction with GDPR, HIPAA and SOC 2 false.