# Anthropic vs DeepSeek

Canonical: https://inetgeek.com/compare/anthropic-vs-deepseek/

Every value below is read from Anthropic's and DeepSeek's own documentation. See https://inetgeek.com/methodology/ for how.

## At a glance

| Criterion | Anthropic | DeepSeek |
| --- | --- | --- |
| Pricing model | Per million tokens, priced separately for input and output and per model, with multipliers stacked on top: prompt caching, a 50% batch discount, a 1.1x uplift for US-only inference on Claude 4.6 and later, and a fast-mode premium. | Per million tokens, with every rate doubling at peak: off-peak is half price, and peak is 01:00–04:00 and 06:00–10:00 UTC on weekdays. Cache hits are priced roughly 30x below cache misses. |
| Input price, top model | $10 per million input tokens for Claude Fable 5.1, the most capable model. Claude Opus 5 is $5 and Claude Sonnet 5 is $2. | $1.32 per million input tokens for deepseek-v4-pro at peak on a cache miss, halving to $0.66 off-peak. A cache hit is $0.044 peak. |
| Output price, top model | $50 per million output tokens for Claude Fable 5.1. Claude Opus 5 is $25 and Claude Sonnet 5 is $10. | $3.96 per million output tokens for deepseek-v4-pro at peak, $1.98 off-peak. |
| Input price, cheapest model | $1 per million input tokens for Claude Haiku 4.5, the cheapest current model, with output at $5. | $0.44 per million input tokens for deepseek-v4-flash at peak on a cache miss, $0.22 off-peak, with output at $1.32 peak. |
| Context window | 1M tokens on Claude Fable 5.1, Claude Opus 5 and Claude Sonnet 5; 200K on Claude Haiku 4.5. Anthropic states the full 1M window is billed at standard per-token rates. | 1M tokens on every model in the current line. |
| Max output tokens | 128K tokens on the top three models, 64K on Claude Haiku 4.5. The Batch API supports up to 300K output tokens on several models behind a beta header. | 384K tokens maximum. |
| Prompt caching | Supported | Supported |

## Where they differ

### Pricing model

- Anthropic: Per million tokens, priced separately for input and output and per model, with multipliers stacked on top: prompt caching, a 50% batch discount, a 1.1x uplift for US-only inference on Claude 4.6 and later, and a fast-mode premium. ([source](https://docs.claude.com/en/docs/about-claude/pricing))
- DeepSeek: Per million tokens, with every rate doubling at peak: off-peak is half price, and peak is 01:00–04:00 and 06:00–10:00 UTC on weekdays. Cache hits are priced roughly 30x below cache misses. ([source](https://api-docs.deepseek.com/quick_start/pricing))

### Input price, top model

- Anthropic: $10 per million input tokens for Claude Fable 5.1, the most capable model. Claude Opus 5 is $5 and Claude Sonnet 5 is $2. ([source](https://docs.claude.com/en/docs/about-claude/pricing))
- DeepSeek: $1.32 per million input tokens for deepseek-v4-pro at peak on a cache miss, halving to $0.66 off-peak. A cache hit is $0.044 peak. ([source](https://api-docs.deepseek.com/quick_start/pricing))

### Output price, top model

- Anthropic: $50 per million output tokens for Claude Fable 5.1. Claude Opus 5 is $25 and Claude Sonnet 5 is $10. ([source](https://docs.claude.com/en/docs/about-claude/pricing))
- DeepSeek: $3.96 per million output tokens for deepseek-v4-pro at peak, $1.98 off-peak. ([source](https://api-docs.deepseek.com/quick_start/pricing))

### Input price, cheapest model

- Anthropic: $1 per million input tokens for Claude Haiku 4.5, the cheapest current model, with output at $5. ([source](https://docs.claude.com/en/docs/about-claude/pricing))
- DeepSeek: $0.44 per million input tokens for deepseek-v4-flash at peak on a cache miss, $0.22 off-peak, with output at $1.32 peak. ([source](https://api-docs.deepseek.com/quick_start/pricing))

### Context window

- Anthropic: 1M tokens on Claude Fable 5.1, Claude Opus 5 and Claude Sonnet 5; 200K on Claude Haiku 4.5. Anthropic states the full 1M window is billed at standard per-token rates. ([source](https://docs.claude.com/en/docs/about-claude/models/overview))
- DeepSeek: 1M tokens on every model in the current line. ([source](https://api-docs.deepseek.com/quick_start/pricing))

### Max output tokens

- Anthropic: 128K tokens on the top three models, 64K on Claude Haiku 4.5. The Batch API supports up to 300K output tokens on several models behind a beta header. ([source](https://docs.claude.com/en/docs/about-claude/models/overview))
- DeepSeek: 384K tokens maximum. ([source](https://api-docs.deepseek.com/quick_start/pricing))

## Which should you choose?

Pick Anthropic if Teams that want the full 1M-token window at the standard per-token rate rather than a long-context surcharge, and that can use caching and batch to cut a frontier-model bill roughly in half.

Pick DeepSeek if Batch-shaped work that can run outside weekday mornings UTC, where off-peak halves an already low rate and a cache hit costs about 3% of a cache miss.

Consider something else: Anthropic — Its cheapest model is $1 per million input tokens, several times what the budget tiers at Google, OpenAI and DeepSeek charge, so high-volume simple work is priced against you.

Consider something else: DeepSeek — Capacity is expressed as concurrent requests rather than tokens per minute, which is a different thing to plan against, and the peak window lands on European working hours.

## Documented by only one

- Batch discount: Anthropic: 50% off both input and output tokens through the Batch API, for asynchronous processing.
- Rate limits: DeepSeek: Concurrency rather than requests per minute: 500 concurrent requests on deepseek-v4-pro and 2,500 on deepseek-v4-flash, per account rather than per key. Exceeding it returns HTTP 429, and expansion is free on request.

## Questions this comparison answers

**Should I choose Anthropic or DeepSeek?**

Pick Anthropic if Teams that want the full 1M-token window at the standard per-token rate rather than a long-context surcharge, and that can use caching and batch to cut a frontier-model bill roughly in half.
Pick DeepSeek if Batch-shaped work that can run outside weekday mornings UTC, where off-peak halves an already low rate and a cache hit costs about 3% of a cache miss.