RXed AI News

Your AI intelligence briefing
SCORE 4.9/1017/20 elements Audited 2026-07-10 · RXed table v1.0

DeepSeek

Frontier-grade open-weight models at a tenth of US prices — but it's an engine, not a platform, and the hosted service's China data residency is a hard no for sensitive work.

Frontier labChatCodingResearchFreemium
Vendor
DeepSeek (Hangzhou DeepSeek Artificial Intelligence, backed by High-Flyer) · www.deepseek.com
Origin
CN — Hangzhou
Pricing
Chat app free · API pay-per-token: V4-Flash ~€0.12/€0.24 per 1M in/out (~€0.0024 cache-hit), V4-Pro ~€0.37/€0.75 · Open weights (MIT) free to self-host
Users (official only)
Not disclosed by vendor; third-party tracker AICPB estimates ~139M app MAU (source, 2026-04-30)

Element scores

Reactive
Retrieval & Memory
Orchestration
Validation
Models
Primitives
Pr6
Prompts
Em2
Embeddings
Cx8.5
Context
Tr2
Tracing
Lg9
LLM
Compositions
Fc6.5
Function calling
Vx
Vector store
Rg4
RAG
Gr2
Guardrails
Mm2.5
Multimodal
Deployment
Ag5.5
Agents
Ft6
Fine-tuning
Fw4
Frameworks & harnesses
Ev2
Evaluations
Sm8
Small models
Emerging
Ma
Multi-agent
Sy
Synthetic data
Pc2
Protocols
In4.5
Interpretability
Th8.5
Thinking models
Tap or hover any element to see why it got that score.

Strengths

Frontier-quality models under MIT license at roughly a tenth of US API prices: V4-Pro (1.6T MoE, 49B active) and V4-Flash both ship a native 1M-token context with sparse attention, thinking modes and tool calls, plus disk-backed context caching that drops cache-hit input to $0.0028/1M. The R1-Distill ladder scales down to a single GPU, and OpenAI- plus Anthropic-format endpoints mean it drops into existing harnesses — including Claude Code — with a base-URL change.

Honest dings

The hosted app and API store prompts in China under a legal framework you cannot contract around — South Korea's regulator documented undisclosed transfers, and US federal agencies, multiple US states, Australia and Taiwan have banned it from government devices. The platform around the models is bare: no embeddings, vector store, fine-tuning service, evals, moderation API or MCP. Models are text-only, and model names shift under you (deepseek-chat and deepseek-reasoner retire 24/07/2026).

Best for: Cost-sensitive builders and self-hosters who bring their own platform: run the MIT weights on EU/US infrastructure for frontier-class reasoning without the hosted-service data risk — and keep sensitive work away from the official app and API.

Sources

Every audit lists the research it rests on — transparency and traceability are the product. Tools evolve: each audit is a snapshot of its audit date, and re-audits supersede older versions (kept below for reference).