DeepSeek
Open-weight reasoning models at a fraction of frontier pricing.
The price-performance outlier. If cost per token is the constraint, nothing else is close.
Wrong forAnyone whose data governance rules out the hosted version.
- Best for
- Strong reasoning on a tight budget
- Pricing
- Free web/app · API pay-as-you-go, token-metered
Verified facts
- Free tier
- No free tier for the API, pay-as-you-go from the first token; the consumer chat web app and mobile app remain free to use
- Entry paid plan
- No subscription tier; API access starts as low as $0.007/1M input tokens (V4 Flash, cache hit, off-peak) and scales up from there
- Bills in
- Token-metered API with a 4-way price matrix (cache hit vs miss, off-peak vs peak, peak runs 2x off-peak): V4 Flash runs $0.007 to $0.44 per 1M input tokens and $0.66 to $1.32 per 1M output tokens; V4 Pro runs $0.022 to $1.32 per 1M input tokens and $1.98 to $3.96 per 1M output tokens
Verified · source: api-docs.deepseek.com
2026-08-19 facts pass corrects the tools.json pricing field, which quoted a single stale 'from $0.28 per million tokens' figure: the current API pricing table has moved to two models (V4 Flash and V4 Pro) with a 4-way price matrix by cache-hit/miss and off-peak/peak, none of which land at $0.28. One third-party aggregator (deepseek.ai/pricing) cites figures roughly 35-60% lower than the vendor's own docs table and is treated as unreliable/stale. No free tier is documented for the API itself, pure pay-as-you-go from the first token; the chat web/app being free is assumed but not independently reverified this pass.
Head to head
What it's actually used for
- Open weights, so it can be self-hosted where data cannot leave the building
- API pricing roughly an order of magnitude below frontier models
We earn nothing from this link. We are not in this tool's affiliate program.
