# DeepSeek

> MIT-licensed open models at some of the lowest first-party prices anywhere.

Canonical: https://www.subconscious.dev/providers/deepseek · By The Subconscious Team · Updated September 30, 2026

- Founded: 2023
- Example models: DeepSeek V4.1 Flash, DeepSeek V4 Pro
- Website: https://www.deepseek.com

## Overview

DeepSeek is the Chinese lab whose open-weight models reset price expectations for the whole market. Its API now serves two models, both with 1M context and 384K max output. V4.1 Flash shipped September 10, 2026 with built-in image understanding at $0.30 in and $1.20 out at peak. V4 Pro, generally available since August 13, costs $1.32 in and $3.96 out at peak. Cache hits cost a few cents per million or less, and the weights ship on Hugging Face under an MIT license.

Since August 16, 2026, DeepSeek bills peak and off-peak rates. Peak runs 01:00 to 04:00 and 06:00 to 10:00 UTC on weekdays, and every other hour costs exactly half, so a US team working normal business hours pays the lower rate. The switch was a real increase, though: V4 Flash output went from a flat $0.28 to $0.66 off-peak. Low, high and max reasoning effort settings give developers a direct lever on output tokens. Because the weights are open, most hosts in this doc also serve DeepSeek, often below DeepSeek's own list.

## Upsides

- Strong open-weight models at some of the lowest first-party prices anywhere.
- Very cheap cache hits, a big win for agents that reread long prefixes.

## Core use cases

- Cost-sensitive agents and batch work scheduled into off-peak windows.
- Self-hosting or fine-tuning a top open model with permissive licensing.

## Downsides

- Data on the hosted API is stored in China, a hard stop for many enterprises.
- Frequent model retirements and repricing force teams to keep re-checking cost models.

## At a glance

| Attribute | Value |
|---|---|
| Model access | Open weights (MIT) |
| Flagship models | DeepSeek V4.1 Flash, V4 Pro |
| Speed | ~35 tok/s on V4 Pro |
| Price | Off-peak hours at half price |
| Customization | Open weights to fine-tune |
| Deployment | First-party API, Hugging Face weights |
| Long context | 1M, 384K max output |

## FAQ

### What is DeepSeek?

DeepSeek is the Chinese lab whose open-weight models reset price expectations for the whole market. Its API now serves two models, both with 1M context and 384K max output. V4.1 Flash shipped September 10, 2026 with built-in image understanding at $0.30 in and $1.20 out at peak. V4 Pro, generally available since August 13, costs $1.32 in and $3.96 out at peak. Cache hits cost a few cents per million or less, and the weights ship on Hugging Face under an MIT license.

### What is DeepSeek best for?

Cost-sensitive agents and batch work scheduled into off-peak windows; Self-hosting or fine-tuning a top open model with permissive licensing.

### How much does DeepSeek cost?

DeepSeek pricing at a glance: Off-peak hours at half price. Rates change often, so check DeepSeek's pricing page before committing.

### How much context does DeepSeek support?

DeepSeek's long-context support: 1M, 384K max output.

### What are the downsides of DeepSeek?

Data on the hosted API is stored in China, a hard stop for many enterprises; Frequent model retirements and repricing force teams to keep re-checking cost models.

### What are the best alternatives to DeepSeek?

Common alternatives include Subconscious, OpenAI, Anthropic, Google Vertex AI, Amazon Bedrock. Each has a head-to-head comparison with DeepSeek on this site.

## Comparisons

- [Subconscious vs DeepSeek](https://www.subconscious.dev/compare/subconscious-vs-deepseek.md)
- [OpenAI vs DeepSeek](https://www.subconscious.dev/compare/openai-vs-deepseek.md)
- [Anthropic vs DeepSeek](https://www.subconscious.dev/compare/anthropic-vs-deepseek.md)
- [Google Vertex AI vs DeepSeek](https://www.subconscious.dev/compare/google-vertex-vs-deepseek.md)
- [Amazon Bedrock vs DeepSeek](https://www.subconscious.dev/compare/aws-bedrock-vs-deepseek.md)
- [Together AI vs DeepSeek](https://www.subconscious.dev/compare/together-ai-vs-deepseek.md)
- [Fireworks AI vs DeepSeek](https://www.subconscious.dev/compare/fireworks-vs-deepseek.md)
- [Baseten vs DeepSeek](https://www.subconscious.dev/compare/baseten-vs-deepseek.md)
- [Groq vs DeepSeek](https://www.subconscious.dev/compare/groq-vs-deepseek.md)
- [Cerebras vs DeepSeek](https://www.subconscious.dev/compare/cerebras-vs-deepseek.md)
- [DeepInfra vs DeepSeek](https://www.subconscious.dev/compare/deepinfra-vs-deepseek.md)
- [Modal vs DeepSeek](https://www.subconscious.dev/compare/modal-vs-deepseek.md)
- [xAI vs DeepSeek](https://www.subconscious.dev/compare/xai-vs-deepseek.md)
- [DeepSeek vs Moonshot AI](https://www.subconscious.dev/compare/deepseek-vs-moonshot-ai.md)
- [DeepSeek vs Z.ai](https://www.subconscious.dev/compare/deepseek-vs-z-ai.md)
- [DeepSeek vs Alibaba Cloud](https://www.subconscious.dev/compare/deepseek-vs-alibaba-cloud.md)
- [DeepSeek vs Meta](https://www.subconscious.dev/compare/deepseek-vs-meta.md)
- [DeepSeek vs SambaNova](https://www.subconscious.dev/compare/deepseek-vs-sambanova.md)
- [DeepSeek vs Nebius](https://www.subconscious.dev/compare/deepseek-vs-nebius.md)
- [DeepSeek vs fal](https://www.subconscious.dev/compare/deepseek-vs-fal.md)
- [DeepSeek vs Novita AI](https://www.subconscious.dev/compare/deepseek-vs-novita-ai.md)
- [DeepSeek vs Parasail](https://www.subconscious.dev/compare/deepseek-vs-parasail.md)
- [DeepSeek vs Inference.net](https://www.subconscious.dev/compare/deepseek-vs-inference-net.md)
- [DeepSeek vs GMI Cloud](https://www.subconscious.dev/compare/deepseek-vs-gmi-cloud.md)
- [DeepSeek vs Sail Research](https://www.subconscious.dev/compare/deepseek-vs-sail-research.md)
- [DeepSeek vs Morph](https://www.subconscious.dev/compare/deepseek-vs-morph.md)
- [DeepSeek vs Relace](https://www.subconscious.dev/compare/deepseek-vs-relace.md)
- [DeepSeek vs TypeSafe AI](https://www.subconscious.dev/compare/deepseek-vs-typesafe-ai.md)
- [DeepSeek vs StepFun](https://www.subconscious.dev/compare/deepseek-vs-stepfun.md)
- [DeepSeek vs Runware](https://www.subconscious.dev/compare/deepseek-vs-runware.md)
- [DeepSeek vs StreamLake](https://www.subconscious.dev/compare/deepseek-vs-streamlake.md)
- [DeepSeek vs Wafer](https://www.subconscious.dev/compare/deepseek-vs-wafer.md)
- [DeepSeek vs RunInfra](https://www.subconscious.dev/compare/deepseek-vs-runinfra.md)
- [DeepSeek vs Particle.AI](https://www.subconscious.dev/compare/deepseek-vs-particle-ai.md)

## Sources

- [DeepSeek pricing Sept 2026](https://justinmckelvey.com/blog/deepseek-pricing)
- [DeepSeek repricing, tokenkarma](https://tokenkarma.app/blog/deepseek-v4-pro-ga-peak-off-peak-pricing-live-2026/)
- [V4 Flash review, Vorp Labs](https://vorplabs.com/models/releases/deepseek-v4-flash)

Pricing and model lineups change often; figures are a snapshot.