DeepSeek at a glance
- Model pages in the directory
- 75
- Live on OpenRouter
- 14
- Price range
- $0.02 to $0.66 input, $0.13 to $2.15 output, per million tokens
12 of them served by other hosts
Models priced in the run of 23 Oct 2026
14 priced. Source: OpenRouter models API, 23 Oct 2026
Listed under 2 provider names in the catalogue: deepseek (43), deepseek-ai (20). The list below shows deepseek.
The provider
DeepSeek's calling card is efficiency and openness. Its models deliver strong reasoning and coding at very low cost, and it releases open weights, so teams can self-host. The house style is research-forward and price-disruptive, and it forced the whole market to re-benchmark cost.
Where it fits
Cost-sensitive and privacy-sensitive teams that want strong reasoning and coding, especially those willing to self-host the open weights.
Strengths
- Frontier-class reasoning and coding at disruptive prices
- Open weights you can self-host for full data control
- Research-forward, with detailed public technical reports
Watch for
- China-based hosted API affects residency for regulated buyers
- Support and tooling are lighter than the US labs
- Rapid releases can mean shifting model IDs
Our take
DeepSeek is the openness-plus-value play. If data control matters, the open weights let you self-host and keep everything in house, which no closed US lab can offer. For reasoning and coding on a budget, it belongs on every shortlist.
DeepSeek is the Hangzhou lab that reset the market's assumptions about cost. Backed by the quantitative hedge fund High-Flyer and led by Liang Wenfeng, it delivered frontier-class reasoning and coding at a fraction of the training and inference budget the field assumed was required, and it did it with open weights.
That combination, strong models plus open weights plus disruptive pricing, is the whole story. Teams can self-host the weights for full data control, which sidesteps the residency question that comes with any China-based hosted API. For reasoning and coding on a budget, DeepSeek forced everyone else to re-benchmark.
The catalogue below is every DeepSeek model we track, joined live to pricing, context, and benchmark scores.
