{"slug": "the-90-off-claude-api-resellers-sub2api", "title": "The 90% Off Claude API: Resellers, Sub2API", "summary": "Chinese providers are reselling Anthropic's Claude and OpenAI's Codex API access at up to 90% off official prices by using open-source tools like Sub2API to wrap subscription plans into simulated API endpoints, a practice that raises data privacy and rate-limit concerns. The arbitrage model buys Pro or Max subscriptions in bulk and pools them across concurrent sessions, exploiting the flat-rate subscription versus per-token API pricing asymmetry. Providers such as ddshub.cc and yunwu.api pitch the same discounted access, but users warn that these resellers can read every prompt sent.", "body_md": "# **The 90% Off Claude API: Resellers, Sub2API\n\nA 90% discount on\n\n[Claude](/en/tags/claude/)and Codex API access should be an immediate red flag, yet the resellers keep multiplying and the deals keep getting bolder. Chinese providers are now undercutting Anthropic and OpenAI's official API pricing by an enormous margin, and the mechanism is surprisingly transparent once you look past the marketing.The play relies on open-source tools like Sub2API, which take a standard subscription plan and wrap it into a simulated API endpoint. The business model is pure arbitrage: buy Pro or Max subscriptions in bulk, pool them across concurrent sessions, then resell that aggregate throughput as \"API access.\" Since subscriptions are flat-rate and API access is per-token, the margin is baked into the structure — no magic, just pricing asymmetry.\n\nI've checked out a few of these providers — ddshub.cc, yunwu.api, among others — and they all pitch the same way: a fraction of the official cost, same models\n\nStory tracker · related coverage\n\n[PISIGuard: Keep Your Private Data Out of AI Chat Logs 2h ago](/en/news/4833/)\n\n[Claude's Malicious Code Leak: How an AI Attacked 3 Real Companies 9h ago](/en/news/4798/)\n\n[Wienerdog: memory + self-improving skills for Claude Code/Codex 1d ago](/en/news/4688/)\n\n[Anthropic AI Hacked 3 Orgs During Testing: A Deep Dive 2d ago](/en/news/4577/)\n\n[AI Reasoning: Why It's Right for the Wrong Reasons 2d ago](/en/news/4549/)\n\n[Three organizations got breached in a controlled exercise — and 3d ago](/en/news/4521/)\n\n[Next **Nightcrawler: Running an AI Pentesting Agent on Your Phone** →](/en/news/4851/)\n\n## All Replies （4）\n\nZ\n\nDon't forget data privacy — those discount resellers can read every prompt you send.\n\n0\n\nJ\n\n[@ZenMaster](/en/users/ZenMaster/)That's why I keep those keys for throwaway stuff only. Anything sensitive stays on the official API, period.\n\n0\n\nT\n\nHonestly, caching Claude's output would be a nightmare to keep consistent. Even if they store plain text results, any slight prompt variation or context change makes it useless. And you're right—KV cache is internal. More likely they just batch requests or use some cheap model for lookalike answers. Total guess though.\n\n0\n\nS\n\nHow do they even handle rate limits on those cut-price keys? Seems like it'd break instantly.\n\n0", "url": "https://wpnews.pro/news/the-90-off-claude-api-resellers-sub2api", "canonical_source": "https://promptcube3.com/en/news/4853/", "published_at": "2026-08-03 12:29:25+00:00", "updated_at": "2026-08-03 12:56:51.634025+00:00", "lang": "en", "topics": ["artificial-intelligence", "ai-products", "ai-infrastructure"], "entities": ["Anthropic", "OpenAI", "Claude", "Codex", "Sub2API", "ddshub.cc", "yunwu.api"], "alternates": {"html": "https://wpnews.pro/news/the-90-off-claude-api-resellers-sub2api", "markdown": "https://wpnews.pro/news/the-90-off-claude-api-resellers-sub2api.md", "text": "https://wpnews.pro/news/the-90-off-claude-api-resellers-sub2api.txt", "jsonld": "https://wpnews.pro/news/the-90-off-claude-api-resellers-sub2api.jsonld"}}