Key findings
- Price per task and cache hits draw praise in 25 of 43 sources, and six developers describe moving work from a coding subscription to the API.
- One developer reports more than 40 billion tokens for over US$800 in 30 days, mostly on V4.1 Flash, and another says five dollars of tokens covers about a week of constant use.
- Output bugs, errors, and slow periods are raised in 5 sources, including answers left in reasoning_content in 13 of 45 replays and internal tool-call markup leaking into content.
- DeepSeek's privacy policy places data storage in the People's Republic of China and uses it for training, with an opt-out only where local law grants one, and 2 developers raise the data policy, one of whom avoids the API for it.
- Lindy, the one company account GILD found, moved most managed model traffic to DeepSeek V4 Flash and cut those inference costs about 90%, serving it from Atlas Cloud, so none of it runs on DeepSeek's own API.
Across 43 public accounts of DeepSeek's own API, 39 state a view: 29 are positive, 4 mixed, and 6 negative, and price per task and cache hits draw praise in 25. Developers report long agent sessions costing cents through cache hits and off-peak hours, and output bugs in agent loops. GILD read English-language accounts published from May to October 2026, with DeepSeek's pricing, limits, privacy policy, and terms checked on October 28, 2026.
Why developers choose it
Price per task leads every other reason. Developers cite cache hits at US$0.006 per million tokens on Flash, half-price off-peak hours, and cache hit rates of 99% and more on long tasks, which make all-day agent sessions cost cents.
Several compare it with coding subscriptions whose limits run out in hours and moved that work to the API.
Where it breaks
Agent loops expose output bugs: a stop finish with the whole answer in reasoning_content, internal tool-call markup in content when a request omits tools, and one-sentence replies that end a turn. Developers also report TLS handshake failures and slow periods.
DeepSeek's status page showed 99.47% uptime for the V4.1 Flash API and 99.74% for V4 Pro over its July to October 2026 window, read on October 28, 2026.
Data terms and restrictions
DeepSeek's privacy policy states that it collects, processes, and stores personal data in the People's Republic of China and uses it to train its models, with an opt-out only where local law grants one, and its Open Platform terms are governed by PRC law. One developer avoids the API because DeepSeek stores and trains on prompts.
Governments restrict DeepSeek products: the BBC reported on October 19, 2026 that DHS bars their use across US federal civilian agencies. In January 2025 Wiz found an exposed DeepSeek database holding chat history, API secrets, and log streams.
Methodology
GILD read English-language public accounts published from May 25 to October 24, 2026 on Hacker News, GitHub, Reddit, and named blogs, and counted only people describing their own use of api.deepseek.com, 43 in all, the same records as the DeepSeek API Signals after a second model checked each against its source. Prices, limits, data terms, and status figures come from DeepSeek's pages fetched on October 28, 2026. Accounts of DeepSeek models served by other hosts are described separately and kept out of the counts.
Products in this piece
Sources
- Official documentation
- 6
- Public source
- 8
- DeepSeek API docs: Models and pricing
- DeepSeek API docs: Rate limit and isolation
- DeepSeek API docs: Context caching
- DeepSeek privacy policy
- DeepSeek Open Platform terms of service (effective Apr 29, 2026)
- DeepSeek status page
- Hacker News comment on 40 billion tokens for US$800 in 30 days on the DeepSeek API (Oct 8, 2026)
- Hacker News comment on 99% cache hits on DeepSeek's own API (Oct 2, 2026)
- Hacker News comment comparing a US$20 plan's limits with US$20 of DeepSeek API spend (Oct 17, 2026)
- deepseek-ai/DeepSeek-V3 issue #1673: answers left in reasoning_content with empty content (Sep 24, 2026)
- deepseek-ai/DeepSeek-V3 issue #1678: deepseek-flash emits internal tool-call markup in content (Sep 30, 2026)
- Bruno Škvorc, Lindy Engineering: Migrating from Claude to DeepSeek (Jun 24, 2026)
- BBC News: DeepSeek banned from US government in race with China for AI dominance (Oct 19, 2026)
- Wiz Research: Wiz Research Uncovers Exposed DeepSeek Database Leaking Sensitive Information (Jan 29, 2025)